File
Blob: src/workerd/io/worker.c++
| 1 | // Copyright (c) 2017-2022 Cloudflare, Inc. |
| 2 | // Licensed under the Apache 2.0 license found in the LICENSE file or at: |
| 3 | // https://opensource.org/licenses/Apache-2.0 |
| 4 | |
| 5 | #include "actor-cache.h" |
| 6 | |
| 7 | #include <workerd/api/actor-state.h> |
| 8 | #include <workerd/api/global-scope.h> |
| 9 | #include <workerd/api/sockets.h> |
| 10 | #include <workerd/api/streams/common.h> // for api::StreamEncoding |
| 11 | #include <workerd/io/cdp.capnp.h> |
| 12 | #include <workerd/io/compatibility-date.h> |
| 13 | #include <workerd/io/features.h> |
| 14 | #include <workerd/io/frankenvalue.h> |
| 15 | #include <workerd/io/tracer.h> |
| 16 | #include <workerd/io/wasm-instantiate-shim.embed.h> |
| 17 | #include <workerd/io/worker.h> |
| 18 | #include <workerd/jsg/async-context.h> |
| 19 | #include <workerd/jsg/inspector.h> |
| 20 | #include <workerd/jsg/jsg.h> |
| 21 | #include <workerd/jsg/modules-new.h> |
| 22 | #include <workerd/jsg/script.h> |
| 23 | #include <workerd/jsg/setup.h> |
| 24 | #include <workerd/jsg/util.h> |
| 25 | #include <workerd/rust/jsg/lib.rs.h> |
| 26 | #include <workerd/rust/jsg/v8.rs.h> |
| 27 | #include <workerd/util/autogate.h> |
| 28 | #include <workerd/util/batch-queue.h> |
| 29 | #include <workerd/util/color-util.h> |
| 30 | #include <workerd/util/mimetype.h> |
| 31 | #include <workerd/util/stream-utils.h> |
| 32 | #include <workerd/util/thread-scopes.h> |
| 33 | #include <workerd/util/uuid.h> |
| 34 | #include <workerd/util/xthreadnotifier.h> |
| 35 | |
| 36 | #include <rust/jsg/ffi.h> |
| 37 | #include <v8-inspector.h> |
| 38 | #include <v8-profiler.h> |
| 39 | #include <v8-wasm.h> |
| 40 | |
| 41 | #include <capnp/compat/json.h> |
| 42 | #include <capnp/message.h> |
| 43 | #include <kj/compat/brotli.h> |
| 44 | #include <kj/compat/gzip.h> |
| 45 | #include <kj/encoding.h> |
| 46 | #include <kj/filesystem.h> |
| 47 | #include <kj/map.h> |
| 48 | |
| 49 | #include <cstdint> |
| 50 | #include <ctime> |
| 51 | #include <map> |
| 52 | #include <numeric> |
| 53 | |
| 54 | #if _WIN32 |
| 55 | #include <io.h> |
| 56 | #include <windows.h> |
| 57 | |
| 58 | #include <kj/win32-api-version.h> |
| 59 | #include <kj/windows-sanity.h> |
| 60 | #else |
| 61 | #include <sys/syscall.h> |
| 62 | #include <unistd.h> |
| 63 | #endif |
| 64 | |
| 65 | namespace workerd { |
| 66 | |
| 67 | namespace { |
| 68 | |
| 69 | constexpr kj::StringPtr logLevelToString(LogLevel level) { |
| 70 | switch (level) { |
| 71 | case LogLevel::DEBUG_: |
| 72 | return "debug"; |
| 73 | case LogLevel::INFO: |
| 74 | return "info"; |
| 75 | case LogLevel::LOG: |
| 76 | return "log"; |
| 77 | case LogLevel::WARN: |
| 78 | return "warn"; |
| 79 | case LogLevel::ERROR: |
| 80 | return "error"; |
| 81 | default: |
| 82 | return "log"; |
| 83 | } |
| 84 | } |
| 85 | |
| 86 | void headersToCDP(const kj::HttpHeaders& in, capnp::JsonValue::Builder out) { |
| 87 | std::map<kj::StringPtr, kj::Vector<kj::StringPtr>> inMap; |
| 88 | in.forEach([&](kj::StringPtr name, kj::StringPtr value) { |
| 89 | inMap.try_emplace(name, 1).first->second.add(value); |
| 90 | }); |
| 91 | |
| 92 | auto outObj = out.initObject(inMap.size()); |
| 93 | auto headersPos = 0; |
| 94 | for (auto& entry: inMap) { |
| 95 | auto field = outObj[headersPos++]; |
| 96 | field.setName(entry.first); |
| 97 | |
| 98 | // CDP uses strange header representation where headers with multiple |
| 99 | // values are merged into one newline-delimited string |
| 100 | field.initValue().setString(kj::strArray(entry.second, "\n")); |
| 101 | } |
| 102 | } |
| 103 | |
| 104 | void stackTraceToCDP(jsg::Lock& js, cdp::Runtime::StackTrace::Builder builder) { |
| 105 | // TODO(cleanup): Maybe use V8Inspector::captureStackTrace() which does this for us. However, it |
| 106 | // produces protocol objects in its own format which want to handle their whole serialization |
| 107 | // to JSON. Also, those protocol objects are defined in generated code which we currently don't |
| 108 | // include in our cached V8 build artifacts; we'd need to fix that. But maybe we should really |
| 109 | // be using the V8-generated protocol objects rather than our parallel capnp versions! |
| 110 | |
| 111 | auto stackTrace = v8::StackTrace::CurrentStackTrace(js.v8Isolate, 10); |
| 112 | auto frameCount = stackTrace->GetFrameCount(); |
| 113 | auto callFrames = builder.initCallFrames(frameCount); |
| 114 | for (int i = 0; i < frameCount; i++) { |
| 115 | auto src = stackTrace->GetFrame(js.v8Isolate, i); |
| 116 | auto dest = callFrames[i]; |
| 117 | auto url = src->GetScriptNameOrSourceURL(); |
| 118 | if (!url.IsEmpty()) { |
| 119 | dest.setUrl(kj::str(url)); |
| 120 | } else { |
| 121 | dest.setUrl(""_kj); |
| 122 | } |
| 123 | dest.setScriptId(kj::str(src->GetScriptId())); |
| 124 | auto func = src->GetFunctionName(); |
| 125 | if (!func.IsEmpty()) { |
| 126 | dest.setFunctionName(kj::str(func)); |
| 127 | } else { |
| 128 | dest.setFunctionName(""_kj); |
| 129 | } |
| 130 | // V8 locations are 1-based, but CDP locations are 0-based... oh, well |
| 131 | dest.setLineNumber(src->GetLineNumber() - 1); |
| 132 | dest.setColumnNumber(src->GetColumn() - 1); |
| 133 | } |
| 134 | } |
| 135 | |
| 136 | kj::Own<capnp::JsonCodec> makeCdpJsonCodec() { |
| 137 | auto codec = kj::heap<capnp::JsonCodec>(); |
| 138 | codec->handleByAnnotation<cdp::Command>(); |
| 139 | codec->handleByAnnotation<cdp::Event>(); |
| 140 | return codec; |
| 141 | } |
| 142 | const capnp::JsonCodec& getCdpJsonCodec() { |
| 143 | static const kj::Own<capnp::JsonCodec> codec = makeCdpJsonCodec(); |
| 144 | return *codec; |
| 145 | } |
| 146 | |
| 147 | } // namespace |
| 148 | |
| 149 | // ======================================================================================= |
| 150 | |
| 151 | namespace { |
| 152 | |
| 153 | using ExceptionOrDuration = kj::OneOf<kj::Exception, kj::Duration>; |
| 154 | |
| 155 | // Inform the inspector of an exception thrown. |
| 156 | // |
| 157 | // Passes `source` as the exception's short message. Reconstructs `message` from `exception` if |
| 158 | // `message` is empty. |
| 159 | void sendExceptionToInspector(jsg::Lock& js, |
| 160 | v8_inspector::V8Inspector& inspector, |
| 161 | UncaughtExceptionSource source, |
| 162 | const jsg::JsValue& exception, |
| 163 | jsg::JsMessage message) { |
| 164 | jsg::sendExceptionToInspector(js, inspector, kj::str(source), exception, message); |
| 165 | } |
| 166 | |
| 167 | void addExceptionToTrace(jsg::Lock& js, |
| 168 | IoContext& ioContext, |
| 169 | BaseTracer& tracer, |
| 170 | UncaughtExceptionSource source, |
| 171 | const jsg::JsValue& exception, |
| 172 | const jsg::TypeHandler<Worker::Api::ErrorInterface>& errorTypeHandler) { |
| 173 | if (source == UncaughtExceptionSource::INTERNAL || |
| 174 | source == UncaughtExceptionSource::INTERNAL_ASYNC) { |
| 175 | // Skip redundant intermediate JS->C++ exception reporting. See: IoContext::runImpl(), |
| 176 | // PromiseWrapper::tryUnwrap() |
| 177 | // |
| 178 | // TODO(someday): Arguably it could make sense to store these exceptions off to the side and |
| 179 | // report them only if they don't end up being duplicates of a later exception that has a more |
| 180 | // specific context. This would cover cases where the C++ code that eventually received the |
| 181 | // exception never ended up reporting it. |
| 182 | return; |
| 183 | } |
| 184 | |
| 185 | auto timestamp = ioContext.now(); |
| 186 | Worker::Api::ErrorInterface error; |
| 187 | |
| 188 | if (exception.isObject()) { |
| 189 | error = KJ_REQUIRE_NONNULL(errorTypeHandler.tryUnwrap(js, exception), |
| 190 | "Should always be possible to unwrap error interface from an object."); |
| 191 | } |
| 192 | |
| 193 | kj::String name; |
| 194 | KJ_IF_SOME(n, error.name) { |
| 195 | name = kj::str(n); |
| 196 | } else { |
| 197 | name = kj::str("Error"); |
| 198 | } |
| 199 | kj::String message; |
| 200 | KJ_IF_SOME(m, error.message) { |
| 201 | message = kj::str(m); |
| 202 | } else { |
| 203 | // This doesn't appear to be an Error object. Fall back to stringifying the whole value as |
| 204 | // the message. |
| 205 | if (!js.v8Isolate->IsExecutionTerminating()) { |
| 206 | v8::TryCatch tryCatch(js.v8Isolate); |
| 207 | try { |
| 208 | message = exception.toString(js); |
| 209 | } catch (jsg::JsExceptionThrown&) { |
| 210 | // Failed to stringify. |
| 211 | // |
| 212 | // Note that we're intentionally not checking tryCatch.CanContinue() here, because we still |
| 213 | // want to continue even if the isolate has been terminated. |
| 214 | } |
| 215 | } |
| 216 | } |
| 217 | |
| 218 | kj::Maybe<kj::String> stack; |
| 219 | KJ_IF_SOME(s, error.stack) { |
| 220 | kj::StringPtr slice = s; |
| 221 | |
| 222 | // Normally `error.stack` repeats the error type and message first. We don't want send two |
| 223 | // copies of that to the trace so we'll strip it off. |
| 224 | if (slice.startsWith(name)) { |
| 225 | slice = slice.slice(name.size()); |
| 226 | if (slice.startsWith(": "_kj)) { |
| 227 | slice = slice.slice(2); |
| 228 | } |
| 229 | } |
| 230 | |
| 231 | if (slice.startsWith(message)) { |
| 232 | slice = slice.slice(message.size()); |
| 233 | if (slice.startsWith("\n")) { |
| 234 | slice = slice.slice(1); |
| 235 | } |
| 236 | } |
| 237 | |
| 238 | if (slice.size() > 0) { |
| 239 | stack = kj::str(slice); |
| 240 | } |
| 241 | } |
| 242 | |
| 243 | tracer.addException(ioContext.getInvocationSpanContext(), timestamp, kj::mv(name), |
| 244 | kj::mv(message), kj::mv(stack)); |
| 245 | } |
| 246 | |
| 247 | void reportStartupError(kj::StringPtr id, |
| 248 | jsg::Lock& js, |
| 249 | const kj::Maybe<std::unique_ptr<v8_inspector::V8Inspector>>& inspector, |
| 250 | const IsolateLimitEnforcer& limitEnforcer, |
| 251 | ExceptionOrDuration limitErrorOrTime, |
| 252 | v8::TryCatch& catcher, |
| 253 | kj::Maybe<Worker::ValidationErrorReporter&> errorReporter, |
| 254 | kj::Maybe<kj::Exception>& permanentException, |
| 255 | SpanParent parentSpan, |
| 256 | bool isDynamicWorker) { |
| 257 | v8::TryCatch catcher2(js.v8Isolate); |
| 258 | ExceptionOrDuration limitErrorOrTime2 = 0 * kj::NANOSECONDS; |
| 259 | try { |
| 260 | KJ_SWITCH_ONEOF(limitErrorOrTime) { |
| 261 | KJ_CASE_ONEOF(limitError, kj::Exception) { |
| 262 | auto description = jsg::extractTunneledExceptionDescription(limitError.getDescription()); |
| 263 | |
| 264 | auto& ex = permanentException.emplace(kj::mv(limitError)); |
| 265 | KJ_IF_SOME(e, errorReporter) { |
| 266 | e.addError(kj::heapString(description)); |
| 267 | } else KJ_IF_SOME(i, inspector) { |
| 268 | // We want to extend just enough CPU time as is necessary to report the exception |
| 269 | // to the inspector here. 10 milliseconds should be more than enough. |
| 270 | auto limitScope = limitEnforcer.enterLoggingJs(js, limitErrorOrTime2); |
| 271 | jsg::sendExceptionToInspector(js, *i.get(), description); |
| 272 | // When the inspector is active, we don't want to throw here because then the inspector |
| 273 | // won't be able to connect and the developer will never know what happened. |
| 274 | } else { |
| 275 | // We should never get here in production if we've validated scripts before deployment. |
| 276 | KJ_LOG(WARNING, "script startup exceeded resource limits", id, ex); |
| 277 | kj::throwFatalException(ex.clone()); |
| 278 | } |
| 279 | } |
| 280 | KJ_CASE_ONEOF_DEFAULT { |
| 281 | if (catcher.HasCaught()) { |
| 282 | js.withinHandleScope([&] { |
| 283 | auto exception = catcher.Exception(); |
| 284 | |
| 285 | permanentException = js.exceptionToKj(js.v8Ref(exception)); |
| 286 | |
| 287 | KJ_IF_SOME(e, errorReporter) { |
| 288 | auto limitScope = limitEnforcer.enterLoggingJs(js, limitErrorOrTime2); |
| 289 | |
| 290 | kj::Vector<kj::String> lines; |
| 291 | lines.add(kj::str("Uncaught ", |
| 292 | jsg::extractTunneledExceptionDescription( |
| 293 | KJ_ASSERT_NONNULL(permanentException).getDescription()))); |
| 294 | jsg::JsMessage message(catcher.Message()); |
| 295 | message.addJsStackTrace(js, lines); |
| 296 | e.addError(kj::strArray(lines, "\n")); |
| 297 | |
| 298 | } else KJ_IF_SOME(i, inspector) { |
| 299 | auto limitScope = limitEnforcer.enterLoggingJs(js, limitErrorOrTime2); |
| 300 | sendExceptionToInspector(js, *i.get(), UncaughtExceptionSource::INTERNAL, |
| 301 | jsg::JsValue(exception), jsg::JsMessage(catcher.Message())); |
| 302 | // When the inspector is active, we don't want to throw here because then the inspector |
| 303 | // won't be able to connect and the developer will never know what happened. |
| 304 | } else { |
| 305 | // We should never get here in production if we've validated scripts before deployment. |
| 306 | // (unless this is a dynamic worker) |
| 307 | kj::Vector<kj::String> lines; |
| 308 | jsg::JsMessage message(catcher.Message()); |
| 309 | message.addJsStackTrace(js, lines); |
| 310 | auto trace = kj::strArray(lines, "; "); |
| 311 | auto description = KJ_ASSERT_NONNULL(permanentException).getDescription(); |
| 312 | auto span = parentSpan.newChild("script_startup_exception"_kjc); |
| 313 | span.setTag("error"_kjc, true); |
| 314 | span.addLog(kj::systemPreciseCalendarClock().now(), "exception"_kjc, |
| 315 | kj::ConstString( |
| 316 | kj::str("script startup threw exception", id, description, trace))); |
| 317 | if (isDynamicWorker) { |
| 318 | // Rethrow the tunneled JSG exception so it converts back to a JS Error. |
| 319 | kj::throwFatalException(KJ_ASSERT_NONNULL(permanentException).clone()); |
| 320 | } else { |
| 321 | KJ_LOG(ERROR, "script startup threw exception", id, description, trace); |
| 322 | KJ_FAIL_REQUIRE("script startup threw exception"); |
| 323 | } |
| 324 | } |
| 325 | }); |
| 326 | } else { |
| 327 | kj::throwFatalException(permanentException |
| 328 | .emplace(KJ_EXCEPTION(FAILED, |
| 329 | "returned empty handle but didn't throw exception?", id)) |
| 330 | .clone()); |
| 331 | } |
| 332 | } |
| 333 | } |
| 334 | } catch (const jsg::JsExceptionThrown&) { |
| 335 | #define LOG_AND_SET_PERM_EXCEPTION(...) \ |
| 336 | KJ_LOG(ERROR, __VA_ARGS__); \ |
| 337 | if (permanentException == kj::none) { \ |
| 338 | permanentException = KJ_EXCEPTION(FAILED, __VA_ARGS__); \ |
| 339 | } |
| 340 | |
| 341 | KJ_SWITCH_ONEOF(limitErrorOrTime2) { |
| 342 | KJ_CASE_ONEOF(limitError2, kj::Exception) { |
| 343 | // TODO(cleanup): If we see this error show up in production, stop logging it, because I |
| 344 | // guess it's not necessarily an error? The other two cases below are more worrying though. |
| 345 | KJ_LOG(ERROR, limitError2); |
| 346 | if (permanentException == kj::none) { |
| 347 | permanentException = kj::mv(limitError2); |
| 348 | } |
| 349 | } |
| 350 | KJ_CASE_ONEOF_DEFAULT { |
| 351 | if (catcher2.HasTerminated()) { |
| 352 | LOG_AND_SET_PERM_EXCEPTION( |
| 353 | "script startup threw exception; during our attempt to stringify the exception, " |
| 354 | "the script apparently was terminated for non-resource-limit reasons.", |
| 355 | id); |
| 356 | } else { |
| 357 | LOG_AND_SET_PERM_EXCEPTION( |
| 358 | "script startup threw exception; furthermore, an attempt to stringify the exception " |
| 359 | "threw another exception, which shouldn't be possible?", |
| 360 | id); |
| 361 | } |
| 362 | } |
| 363 | } |
| 364 | #undef LOG_AND_SET_PERM_EXCEPTION |
| 365 | } |
| 366 | } |
| 367 | |
| 368 | uint64_t getCurrentThreadId() { |
| 369 | #if __linux__ |
| 370 | return syscall(SYS_gettid); |
| 371 | #elif _WIN32 |
| 372 | return GetCurrentThreadId(); |
| 373 | #else |
| 374 | // Assume MacOS or BSD |
| 375 | uint64_t tid; |
| 376 | pthread_threadid_np(nullptr, &tid); |
| 377 | return tid; |
| 378 | #endif |
| 379 | } |
| 380 | |
| 381 | } // namespace |
| 382 | |
| 383 | // Represents a thread's attempt to take an async lock. Each Isolate has a linked list of |
| 384 | // `AsyncWaiter`s. A particular thread only ever owns one `AsyncWaiter` at a time. |
| 385 | class Worker::AsyncWaiter: public kj::Refcounted { |
| 386 | public: |
| 387 | AsyncWaiter(kj::Own<const Isolate> isolate); |
| 388 | ~AsyncWaiter() noexcept; |
| 389 | KJ_DISALLOW_COPY_AND_MOVE(AsyncWaiter); |
| 390 | |
| 391 | private: |
| 392 | // Executor for this waiter's thread. |
| 393 | const kj::Executor& executor; |
| 394 | |
| 395 | // The isolate for which this waiter is currently waiting. |
| 396 | kj::Own<const Isolate> isolate; |
| 397 | |
| 398 | // Promise/fulfiller to fire when the waiter reaches the front of the list for the corresponding |
| 399 | // isolate. |
| 400 | kj::ForkedPromise<void> readyPromise = nullptr; |
| 401 | kj::Own<kj::CrossThreadPromiseFulfiller<void>> readyFulfiller; |
| 402 | |
| 403 | // Promise/fulfiller to fire when the AsyncLock is finally released. This is used when a thread |
| 404 | // tries to take locks on multiple different isolates concurrently, in order to serialize the |
| 405 | // locks so only one is taken at a time. This is NOT a cross-thread fulfiller; it can only be |
| 406 | // fulfilled by the thread that owns the waiter. |
| 407 | kj::ForkedPromise<void> releasePromise = nullptr; |
| 408 | kj::Own<kj::PromiseFulfiller<void>> releaseFulfiller; |
| 409 | |
| 410 | // Protected by the lock on `Isolate::asyncWaiters` for the isolate identified by |
| 411 | // `currentIsolate`. Must be null if `currentIsolate` is null. (All other members of `Waiter` |
| 412 | // can only be accessed by the thread that created the `Waiter`.) |
| 413 | kj::Maybe<AsyncWaiter&> next; |
| 414 | kj::Maybe<AsyncWaiter&>* prev; |
| 415 | |
| 416 | static const kj::EventLoopLocal<AsyncWaiter*> threadCurrentWaiter; |
| 417 | |
| 418 | friend class Worker::Isolate; |
| 419 | friend class Worker::AsyncLock; |
| 420 | }; |
| 421 | |
| 422 | class Worker::InspectorClient: public v8_inspector::V8InspectorClient { |
| 423 | public: |
| 424 | // Wall time in milliseconds with millisecond precision. console.time() and friends rely on this |
| 425 | // function to implement timers. |
| 426 | double currentTimeMS() override { |
| 427 | auto timePoint = kj::UNIX_EPOCH; |
| 428 | |
| 429 | KJ_IF_SOME(ioContext, IoContext::tryCurrent()) { |
| 430 | // We're on a request-serving thread. |
| 431 | timePoint = ioContext.now(); |
| 432 | } else { |
| 433 | auto lockedState = state.lockExclusive(); |
| 434 | KJ_IF_SOME(info, lockedState->inspectorTimerInfo) { |
| 435 | if (info.threadId == getCurrentThreadId()) { |
| 436 | // We're on an inspector-serving thread. |
| 437 | timePoint = |
| 438 | info.timer.now() + info.timerOffset - kj::origin<kj::TimePoint>() + kj::UNIX_EPOCH; |
| 439 | } |
| 440 | } |
| 441 | // We're at script startup time -- just return the Epoch. |
| 442 | } |
| 443 | return (timePoint - kj::UNIX_EPOCH) / kj::MILLISECONDS; |
| 444 | } |
| 445 | |
| 446 | void setInspectorTimerInfo(kj::Timer& timer, kj::Duration timerOffset) { |
| 447 | auto lockedState = state.lockExclusive(); |
| 448 | lockedState->inspectorTimerInfo = InspectorTimerInfo{timer, timerOffset, getCurrentThreadId()}; |
| 449 | } |
| 450 | |
| 451 | void setChannel(Worker::Isolate::InspectorChannelImpl& channel) { |
| 452 | auto lockedState = state.lockExclusive(); |
| 453 | // There is only one active inspector channel at a time in workerd. The teardown of any |
| 454 | // previous channel should have invalidated `lockedState->channel`. |
| 455 | KJ_REQUIRE(lockedState->channel == kj::none); |
| 456 | lockedState->channel = channel; |
| 457 | } |
| 458 | |
| 459 | void resetChannel() { |
| 460 | auto lockedState = state.lockExclusive(); |
| 461 | lockedState->channel = kj::none; |
| 462 | } |
| 463 | |
| 464 | // This method is called by v8 when a breakpoint or debugger statement is hit. This method |
| 465 | // processes debugger messages until `Debugger.resume()` is called, when v8 then calls |
| 466 | // `quitMessageLoopOnPause()`. |
| 467 | // |
| 468 | // This method is ultimately called from the `InspectorChannelImpl` and the isolate lock is |
| 469 | // held when this method is called. |
| 470 | void runMessageLoopOnPause(int contextGroupId) override { |
| 471 | auto lockedState = state.lockExclusive(); |
| 472 | KJ_IF_SOME(channel, lockedState->channel) { |
| 473 | runMessageLoop = true; |
| 474 | do { |
| 475 | if (!dispatchOneMessageDuringPause(channel)) { |
| 476 | break; |
| 477 | } |
| 478 | } while (runMessageLoop); |
| 479 | } |
| 480 | } |
| 481 | |
| 482 | // This method is called by v8 to resume execution after a breakpoint is hit. |
| 483 | void quitMessageLoopOnPause() override { |
| 484 | runMessageLoop = false; |
| 485 | } |
| 486 | |
| 487 | private: |
| 488 | static bool dispatchOneMessageDuringPause(Worker::Isolate::InspectorChannelImpl& channel); |
| 489 | |
| 490 | struct InspectorTimerInfo { |
| 491 | kj::Timer& timer; |
| 492 | kj::Duration timerOffset; |
| 493 | uint64_t threadId; |
| 494 | }; |
| 495 | |
| 496 | bool runMessageLoop; |
| 497 | |
| 498 | // State that may be set on a thread other than the isolate thread. |
| 499 | // These are typically set in attachInspector when an inspector connection is |
| 500 | // made. |
| 501 | struct State { |
| 502 | // Inspector channel to use to pump messages. |
| 503 | kj::Maybe<Worker::Isolate::InspectorChannelImpl&> channel; |
| 504 | |
| 505 | // The timer and offset for the inspector-serving thread. |
| 506 | kj::Maybe<InspectorTimerInfo> inspectorTimerInfo; |
| 507 | }; |
| 508 | kj::MutexGuarded<State> state; |
| 509 | }; |
| 510 | |
| 511 | static thread_local const Worker::Api* currentApi = nullptr; |
| 512 | |
| 513 | const Worker::Api& Worker::Api::current() { |
| 514 | KJ_REQUIRE(currentApi != nullptr, "not running JavaScript"); |
| 515 | return *currentApi; |
| 516 | } |
| 517 | |
| 518 | kj::Maybe<const Worker::Api&> Worker::Api::tryCurrent() { |
| 519 | if (currentApi != nullptr) { |
| 520 | return *currentApi; |
| 521 | } |
| 522 | return kj::none; |
| 523 | } |
| 524 | |
| 525 | jsg::Optional<jsg::Ref<api::CacheContext>> Worker::Api::getCtxCacheProperty(jsg::Lock& js) const { |
| 526 | return kj::none; |
| 527 | } |
| 528 | |
| 529 | struct Worker::Impl { |
| 530 | kj::Maybe<jsg::JsContext<api::ServiceWorkerGlobalScope>> context; |
| 531 | |
| 532 | // The environment blob to pass to handlers. |
| 533 | kj::Maybe<jsg::Value> env; |
| 534 | kj::Maybe<jsg::Value> ctxExports; |
| 535 | |
| 536 | // Note: The default export is given the string name "default", because that's what V8 tells us, |
| 537 | // and so it's easiest to go with it. I guess that means that you can't actually name an export |
| 538 | // "default"? |
| 539 | kj::HashMap<kj::String, api::ExportedHandler> namedHandlers; |
| 540 | kj::HashMap<kj::String, ActorClassInfo> actorClasses; |
| 541 | kj::HashMap<kj::String, EntrypointClass> statelessClasses; |
| 542 | kj::HashMap<kj::String, EntrypointClass> workflowClasses; |
| 543 | |
| 544 | // If set, then any attempt to use this worker shall throw this exception. |
| 545 | kj::Maybe<kj::Exception> permanentException; |
| 546 | }; |
| 547 | |
| 548 | // Note that Isolate mutable state is protected by locking the JsgWorkerIsolate unless otherwise |
| 549 | // noted. |
| 550 | struct Worker::Isolate::Impl { |
| 551 | IsolateObserver& metrics; |
| 552 | kj::Own<InspectorClient> inspectorClient; |
| 553 | kj::Maybe<std::unique_ptr<v8_inspector::V8Inspector>> inspector; |
| 554 | InspectorPolicy inspectorPolicy; |
| 555 | kj::Maybe<kj::Own<v8::CpuProfiler>> profiler; |
| 556 | ActorCache::SharedLru actorCacheLru; |
| 557 | |
| 558 | // Used by JSG/Rust integration. |
| 559 | ::rust::Box<::workerd::rust::jsg::Realm> realm; |
| 560 | |
| 561 | // UUID for this isolate, initialized first time getUuid() is called. |
| 562 | kj::Lazy<kj::String> uuid; |
| 563 | |
| 564 | // Notification messages to deliver to the next inspector client when it connects. |
| 565 | kj::Vector<kj::String> queuedNotifications; |
| 566 | |
| 567 | // Set of warning log lines that should not be logged to the inspector again. |
| 568 | kj::HashSet<kj::String> warningOnceDescriptions; |
| 569 | |
| 570 | // Set of error log lines that should not be logged again. |
| 571 | kj::HashSet<kj::String> errorOnceDescriptions; |
| 572 | |
| 573 | // Instantaneous count of how many threads are trying to or have successfully obtained an |
| 574 | // AsyncLock on this isolate, used to implement getCurrentLoad(). |
| 575 | mutable uint lockAttemptGauge = 0; |
| 576 | |
| 577 | // Atomically incremented upon every successful lock. The ThreadProgressCounter in Impl::Lock |
| 578 | // registers a reference to `lockSuccessCounter` as the thread's progress counter during a lock |
| 579 | // attempt. This allows watchdogs to see evidence of forward progress in other threads, even if |
| 580 | // their own thread has blocked waiting for the lock for a long time. |
| 581 | mutable uint64_t lockSuccessCount = 0; |
| 582 | |
| 583 | // Wrapper around JsgWorkerIsolate::Lock and various RAII objects which help us report metrics, |
| 584 | // measure instantaneous load, avoid spurious watchdog kills, and defer context destruction. |
| 585 | // |
| 586 | // Always use this wrapper in code which may face lock contention (that's mostly everywhere). |
| 587 | class Lock { |
| 588 | |
| 589 | public: |
| 590 | explicit Lock( |
| 591 | const Worker::Isolate& isolate, Worker::LockType lockType, jsg::V8StackScope& stackScope) |
| 592 | : impl(*isolate.impl), |
| 593 | metrics([&isolate, &lockType]() -> kj::Maybe<kj::Own<IsolateObserver::LockTiming>> { |
| 594 | KJ_SWITCH_ONEOF(lockType.origin) { |
| 595 | KJ_CASE_ONEOF(sync, Worker::Lock::TakeSynchronously) { |
| 596 | // TODO(perf): We could do some tracking here to discover overly harmful synchronous |
| 597 | // locks. |
| 598 | return isolate.getMetrics().tryCreateLockTiming(sync.getRequest()); |
| 599 | } |
| 600 | KJ_CASE_ONEOF(async, AsyncLock*) { |
| 601 | KJ_REQUIRE(async->waiter->isolate.get() == &isolate, |
| 602 | "async lock was taken against a different isolate than the synchronous lock"); |
| 603 | return kj::mv(async->lockTiming); |
| 604 | } |
| 605 | } |
| 606 | KJ_UNREACHABLE; |
| 607 | }()), |
| 608 | progressCounter(impl.lockSuccessCount), |
| 609 | oldCurrentApi(currentApi), |
| 610 | limitEnforcer(isolate.getLimitEnforcer()), |
| 611 | loggingOptions(isolate.loggingOptions), |
| 612 | lock(isolate.api->lock(stackScope)) { |
| 613 | WarnAboutIsolateLockScope::maybeWarn(); |
| 614 | |
| 615 | // Increment the success count to expose forward progress to all threads. |
| 616 | __atomic_add_fetch(&impl.lockSuccessCount, 1, __ATOMIC_RELAXED); |
| 617 | metrics.locked(); |
| 618 | |
| 619 | // We record the current lock so our GC prologue/epilogue callbacks can report GC time via |
| 620 | // Jaeger tracing. |
| 621 | KJ_DASSERT(impl.currentLock == kj::none, "Isolate lock taken recursively"); |
| 622 | impl.currentLock = *this; |
| 623 | |
| 624 | // Now's a good time to destroy any workers queued up for destruction. |
| 625 | auto workersToDestroy = impl.workerDestructionQueue.lockExclusive()->pop(); |
| 626 | for (auto& workerImpl: workersToDestroy.asArrayPtr()) { |
| 627 | KJ_IF_SOME(c, workerImpl->context) { |
| 628 | disposeContext(kj::mv(c)); |
| 629 | } |
| 630 | workerImpl = nullptr; |
| 631 | } |
| 632 | |
| 633 | currentApi = isolate.api.get(); |
| 634 | } |
| 635 | ~Lock() noexcept(false) { |
| 636 | currentApi = oldCurrentApi; |
| 637 | |
| 638 | #ifdef KJ_DEBUG |
| 639 | // We lack a KJ_DASSERT_NONNULL because it would have to look a lot like KJ_IF_SOME, thus |
| 640 | // we use a pragma around KJ_DEBUG here. |
| 641 | auto& implCurrentLock = KJ_ASSERT_NONNULL(impl.currentLock, "Isolate lock released twice"); |
| 642 | KJ_ASSERT(&implCurrentLock == this, "Isolate lock released recursively"); |
| 643 | #endif |
| 644 | |
| 645 | if (shouldReportIsolateMetrics) { |
| 646 | // The isolate asked this lock to report the stats when it released. Let's do it. |
| 647 | limitEnforcer.reportMetrics(impl.metrics); |
| 648 | } |
| 649 | impl.currentLock = kj::none; |
| 650 | } |
| 651 | KJ_DISALLOW_COPY_AND_MOVE(Lock); |
| 652 | |
| 653 | void setupContext(v8::Local<v8::Context> context) { |
| 654 | // The V8Inspector implements the `console` object. |
| 655 | KJ_IF_SOME(i, impl.inspector) { |
| 656 | i.get()->contextCreated( |
| 657 | v8_inspector::V8ContextInfo(context, 1, jsg::toInspectorStringView("Worker"))); |
| 658 | } |
| 659 | Worker::setupContext(*lock, context, loggingOptions); |
| 660 | } |
| 661 | |
| 662 | void disposeContext(jsg::JsContext<api::ServiceWorkerGlobalScope> context) { |
| 663 | lock->withinHandleScope([&] { |
| 664 | auto v8Context = context.getHandle(*lock); |
| 665 | context->clear(); |
| 666 | KJ_IF_SOME(i, impl.inspector) { |
| 667 | i.get()->contextDestroyed(v8Context); |
| 668 | } |
| 669 | { auto drop = kj::mv(context); } |
| 670 | lock->v8Isolate->ContextDisposedNotification(v8::ContextDependants::kNoDependants); |
| 671 | }); |
| 672 | } |
| 673 | |
| 674 | void gcPrologue() { |
| 675 | metrics.gcPrologue(); |
| 676 | // Filter out tracked WASM instance entries where the instance has been |
| 677 | // garbage-collected (weak instanceRef is empty), allowing the linear memory |
| 678 | // to be reclaimed. |
| 679 | limitEnforcer.getTrackedWasmInstances().filter(*lock); |
| 680 | } |
| 681 | void gcEpilogue() { |
| 682 | metrics.gcEpilogue(); |
| 683 | } |
| 684 | |
| 685 | // Call limitEnforcer.exitJs(), and also schedule to call limitEnforcer.reportMetrics() |
| 686 | // later. Returns true if condemned. We take a mutable reference to it to make sure the caller |
| 687 | // believes it has exclusive access. |
| 688 | bool checkInWithLimitEnforcer(Worker::Isolate& isolate); |
| 689 | |
| 690 | private: |
| 691 | const Impl& impl; |
| 692 | IsolateObserver::LockRecord metrics; |
| 693 | ThreadProgressCounter progressCounter; |
| 694 | bool shouldReportIsolateMetrics = false; |
| 695 | const Api* oldCurrentApi; |
| 696 | |
| 697 | const IsolateLimitEnforcer& limitEnforcer; // only so we can call getIsolateStats() |
| 698 | |
| 699 | // When structuredLogging is YES AND consoleMode is STDOUT js logs will be emitted to STDOUT |
| 700 | // as newline separated json objects |
| 701 | LoggingOptions loggingOptions; |
| 702 | |
| 703 | public: |
| 704 | kj::Own<jsg::Lock> lock; |
| 705 | }; |
| 706 | |
| 707 | // Protected by v8::Locker -- if v8::Locker::IsLocked(isolate) is true, then it is safe to access |
| 708 | // this variable. |
| 709 | mutable kj::Maybe<Lock&> currentLock; |
| 710 | |
| 711 | static constexpr auto WORKER_DESTRUCTION_QUEUE_INITIAL_SIZE = 8; |
| 712 | static constexpr auto WORKER_DESTRUCTION_QUEUE_MAX_CAPACITY = 100; |
| 713 | |
| 714 | // Similar in spirit to the deferred destruction queue in jsg::IsolateBase. When a Worker is |
| 715 | // destroyed, it puts its Impl, which contains objects that need to be destroyed under the isolate |
| 716 | // lock, into this queue. Our own Isolate::Impl::Lock implementation then clears this queue the |
| 717 | // next time the isolate is locked, whether that be by a connection thread, or the Worker's own |
| 718 | // destructor if it owns the last `kj::Own<const Script>` reference. |
| 719 | // |
| 720 | // Fairly obviously, this member is protected by its own mutex, not the isolate lock. |
| 721 | const kj::MutexGuarded<BatchQueue<kj::Own<Worker::Impl>>> workerDestructionQueue{ |
| 722 | WORKER_DESTRUCTION_QUEUE_INITIAL_SIZE, WORKER_DESTRUCTION_QUEUE_MAX_CAPACITY}; |
| 723 | // TODO(cleanup): The only reason this exists and we can't just rely on the isolate's regular |
| 724 | // deferred destruction queue to lazily destroy the various V8 objects in Worker::Impl is |
| 725 | // because our GlobalScope object needs to have a function called on it, and any attached |
| 726 | // inspector needs to be notified. JSG doesn't know about these things. |
| 727 | |
| 728 | struct IsolateState { |
| 729 | kj::Own<InspectorClient> inspectorClient; |
| 730 | kj::Maybe<std::unique_ptr<v8_inspector::V8Inspector>> inspector; |
| 731 | ::rust::Box<::workerd::rust::jsg::Realm> realm; |
| 732 | }; |
| 733 | |
| 734 | static IsolateState initIsolate( |
| 735 | const Api& api, IsolateLimitEnforcer& limitEnforcer, InspectorPolicy inspectorPolicy) { |
| 736 | auto inspectorClient = kj::heap<InspectorClient>(); |
| 737 | // Default constructor of ::rust::Box is deleted, so we use a Maybe to delay initialization. |
| 738 | kj::Maybe<::rust::Box<::workerd::rust::jsg::Realm>> realm; |
| 739 | kj::Maybe<std::unique_ptr<v8_inspector::V8Inspector>> inspector; |
| 740 | jsg::runInV8Stack([&](jsg::V8StackScope& stackScope) { |
| 741 | auto lock = api.lock(stackScope); |
| 742 | auto featureFlagsWords = capnp::canonicalize(api.getFeatureFlags()); |
| 743 | realm = ::workerd::rust::jsg::realm_create( |
| 744 | lock->v8Isolate, featureFlagsWords.asBytes().as<kj_rs::Rust>()); |
| 745 | lock->v8Isolate->SetData( |
| 746 | ::workerd::jsg::SetDataIndex::SET_DATA_RUST_REALM, &*KJ_REQUIRE_NONNULL(realm)); |
| 747 | |
| 748 | limitEnforcer.customizeIsolate(lock->v8Isolate); |
| 749 | if (inspectorPolicy != InspectorPolicy::DISALLOW) { |
| 750 | // We just created our isolate, so we don't need to use Isolate::Impl::Lock. |
| 751 | KJ_ASSERT(!isMultiTenantProcess(), "inspector is not safe in multi-tenant processes"); |
| 752 | inspector = v8_inspector::V8Inspector::create(lock->v8Isolate, inspectorClient.get()); |
| 753 | } |
| 754 | }); |
| 755 | return {kj::mv(inspectorClient), kj::mv(inspector), kj::mv(KJ_REQUIRE_NONNULL(realm))}; |
| 756 | } |
| 757 | |
| 758 | Impl(IsolateObserver& metrics, |
| 759 | IsolateLimitEnforcer& limitEnforcer, |
| 760 | InspectorPolicy inspectorPolicy, |
| 761 | IsolateState state) |
| 762 | : metrics(metrics), |
| 763 | inspectorClient(kj::mv(state.inspectorClient)), |
| 764 | inspector(kj::mv(state.inspector)), |
| 765 | inspectorPolicy(inspectorPolicy), |
| 766 | actorCacheLru(limitEnforcer.getActorCacheLruOptions()), |
| 767 | realm(kj::mv(state.realm)) {} |
| 768 | |
| 769 | Impl(const Api& api, |
| 770 | IsolateObserver& metrics, |
| 771 | IsolateLimitEnforcer& limitEnforcer, |
| 772 | InspectorPolicy inspectorPolicy) |
| 773 | : Impl(metrics, |
| 774 | limitEnforcer, |
| 775 | inspectorPolicy, |
| 776 | initIsolate(api, limitEnforcer, inspectorPolicy)) {} |
| 777 | }; |
| 778 | |
| 779 | namespace { |
| 780 | |
| 781 | class CpuProfilerDisposer final: public kj::Disposer { |
| 782 | public: |
| 783 | virtual void disposeImpl(void* pointer) const override { |
| 784 | reinterpret_cast<v8::CpuProfiler*>(pointer)->Dispose(); |
| 785 | } |
| 786 | |
| 787 | static const CpuProfilerDisposer instance; |
| 788 | }; |
| 789 | |
| 790 | const CpuProfilerDisposer CpuProfilerDisposer::instance{}; |
| 791 | |
| 792 | static constexpr kj::StringPtr PROFILE_NAME = "Default Profile"_kj; |
| 793 | |
| 794 | static void setSamplingInterval(v8::CpuProfiler& profiler, int interval) { |
| 795 | profiler.SetSamplingInterval(interval); |
| 796 | } |
| 797 | |
| 798 | static void startProfiling(jsg::Lock& js, v8::CpuProfiler& profiler) { |
| 799 | js.withinHandleScope([&] { |
| 800 | v8::CpuProfilingOptions options( |
| 801 | v8::kLeafNodeLineNumbers, v8::CpuProfilingOptions::kNoSampleLimit); |
| 802 | profiler.StartProfiling(jsg::v8StrIntern(js.v8Isolate, PROFILE_NAME), kj::mv(options)); |
| 803 | }); |
| 804 | } |
| 805 | |
| 806 | static void stopProfiling(jsg::Lock& js, v8::CpuProfiler& profiler, cdp::Command::Builder& cmd) { |
| 807 | js.withinHandleScope([&] { |
| 808 | auto cpuProfile = profiler.StopProfiling(jsg::v8StrIntern(js.v8Isolate, PROFILE_NAME)); |
| 809 | if (cpuProfile == nullptr) return; // profiling never started |
| 810 | |
| 811 | kj::Vector<const v8::CpuProfileNode*> allNodes; |
| 812 | kj::Vector<const v8::CpuProfileNode*> unvisited; |
| 813 | |
| 814 | unvisited.add(cpuProfile->GetTopDownRoot()); |
| 815 | while (!unvisited.empty()) { |
| 816 | auto next = unvisited.back(); |
| 817 | allNodes.add(next); |
| 818 | unvisited.removeLast(); |
| 819 | for (int i = 0; i < next->GetChildrenCount(); i++) { |
| 820 | unvisited.add(next->GetChild(i)); |
| 821 | } |
| 822 | } |
| 823 | |
| 824 | auto res = cmd.getProfilerStop().initResult(); |
| 825 | auto profile = res.initProfile(); |
| 826 | profile.setStartTime(cpuProfile->GetStartTime()); |
| 827 | profile.setEndTime(cpuProfile->GetEndTime()); |
| 828 | |
| 829 | auto nodes = profile.initNodes(allNodes.size()); |
| 830 | for (auto i: kj::indices(allNodes)) { |
| 831 | auto nodeBuilder = nodes[i]; |
| 832 | nodeBuilder.setId(allNodes[i]->GetNodeId()); |
| 833 | |
| 834 | auto callFrame = nodeBuilder.initCallFrame(); |
| 835 | callFrame.setFunctionName(allNodes[i]->GetFunctionNameStr()); |
| 836 | callFrame.setScriptId(kj::str(allNodes[i]->GetScriptId())); |
| 837 | callFrame.setUrl(allNodes[i]->GetScriptResourceNameStr()); |
| 838 | // V8 locations are 1-based, but CDP locations are 0-based... |
| 839 | callFrame.setLineNumber(allNodes[i]->GetLineNumber() - 1); |
| 840 | callFrame.setColumnNumber(allNodes[i]->GetColumnNumber() - 1); |
| 841 | |
| 842 | nodeBuilder.setHitCount(allNodes[i]->GetHitCount()); |
| 843 | |
| 844 | auto children = nodeBuilder.initChildren(allNodes[i]->GetChildrenCount()); |
| 845 | for (int j = 0; j < allNodes[i]->GetChildrenCount(); j++) { |
| 846 | children.set(j, allNodes[i]->GetChild(j)->GetNodeId()); |
| 847 | } |
| 848 | |
| 849 | auto hitLineCount = allNodes[i]->GetHitLineCount(); |
| 850 | auto lineBuffer = kj::heapArray<v8::CpuProfileNode::LineTick>(hitLineCount); |
| 851 | allNodes[i]->GetLineTicks(lineBuffer.begin(), lineBuffer.size()); |
| 852 | |
| 853 | auto positionTicks = nodeBuilder.initPositionTicks(hitLineCount); |
| 854 | for (uint j = 0; j < hitLineCount; j++) { |
| 855 | auto positionTick = positionTicks[j]; |
| 856 | positionTick.setLine(lineBuffer[j].line); |
| 857 | positionTick.setTicks(lineBuffer[j].hit_count); |
| 858 | } |
| 859 | } |
| 860 | |
| 861 | auto sampleCount = cpuProfile->GetSamplesCount(); |
| 862 | auto samples = profile.initSamples(sampleCount); |
| 863 | auto timeDeltas = profile.initTimeDeltas(sampleCount); |
| 864 | auto lastTimestamp = cpuProfile->GetStartTime(); |
| 865 | for (int i = 0; i < sampleCount; i++) { |
| 866 | samples.set(i, cpuProfile->GetSample(i)->GetNodeId()); |
| 867 | auto sampleTime = cpuProfile->GetSampleTimestamp(i); |
| 868 | timeDeltas.set(i, sampleTime - lastTimestamp); |
| 869 | lastTimestamp = sampleTime; |
| 870 | } |
| 871 | }); |
| 872 | } |
| 873 | |
| 874 | } // anonymous namespace |
| 875 | |
| 876 | struct Worker::Script::Impl { |
| 877 | kj::Own<workerd::VirtualFileSystem> vfs; |
| 878 | kj::Maybe<kj::Arc<workerd::jsg::modules::ModuleRegistry>> maybeNewModuleRegistry; |
| 879 | // When using the new module registry, the module registry itself holds the |
| 880 | // SchemaLoader, so we don't need to hold it here. When using the original |
| 881 | // module registry, however, we need a schema loader to instantiate capnp |
| 882 | // modules and bindings. |
| 883 | kj::Maybe<kj::Own<capnp::SchemaLoader>> maybeSchemaLoader; |
| 884 | |
| 885 | kj::OneOf<jsg::NonModuleScript, kj::Path> unboundScriptOrMainModule; |
| 886 | |
| 887 | kj::Array<CompiledGlobal> globals; |
| 888 | |
| 889 | kj::Maybe<jsg::JsContext<api::ServiceWorkerGlobalScope>> moduleContext; |
| 890 | |
| 891 | // If set, then any attempt to use this script shall throw this exception. |
| 892 | kj::Maybe<kj::Exception> permanentException; |
| 893 | |
| 894 | Impl(kj::Own<workerd::VirtualFileSystem> vfs, |
| 895 | kj::Maybe<kj::Arc<workerd::jsg::modules::ModuleRegistry>> maybeNewModuleRegistry) |
| 896 | : vfs(kj::mv(vfs)), |
| 897 | maybeNewModuleRegistry(kj::mv(maybeNewModuleRegistry)) { |
| 898 | if (this->maybeNewModuleRegistry == kj::none) { |
| 899 | maybeSchemaLoader = kj::heap<capnp::SchemaLoader>(); |
| 900 | } |
| 901 | } |
| 902 | |
| 903 | struct DynamicImportResult { |
| 904 | jsg::Value value; |
| 905 | bool isException = false; |
| 906 | DynamicImportResult(jsg::Value value, bool isException = false) |
| 907 | : value(kj::mv(value)), |
| 908 | isException(isException) {} |
| 909 | }; |
| 910 | using DynamicImportHandler = kj::Function<jsg::Value()>; |
| 911 | |
| 912 | void configureDynamicImports(jsg::Lock& js, jsg::ModuleRegistry& modules) { |
| 913 | // This is only used with the original module registry implementation. |
| 914 | KJ_ASSERT(!FeatureFlags::get(js).getNewModuleRegistry(), |
| 915 | "legacy dynamic imports must not be used with the new module registry"); |
| 916 | static auto constexpr handleDynamicImport = |
| 917 | [](kj::Own<const Worker> worker, DynamicImportHandler handler, |
| 918 | kj::Maybe<jsg::Ref<jsg::AsyncContextFrame>> asyncContext) |
| 919 | -> kj::Promise<DynamicImportResult> { |
| 920 | co_await kj::yield(); |
| 921 | auto asyncLock = co_await worker->takeAsyncLockWithoutRequest(nullptr); |
| 922 | |
| 923 | co_return worker->runInLockScope(asyncLock, [&](Worker::Lock& lock) { |
| 924 | TmpDirStoreScope tmpDirStoreScope; |
| 925 | return JSG_WITHIN_CONTEXT_SCOPE(lock, lock.getContext(), [&](jsg::Lock& js) { |
| 926 | jsg::AsyncContextFrame::Scope asyncContextScope(js, asyncContext); |
| 927 | |
| 928 | // We have to wrap the call to handler in a try catch here because |
| 929 | // we have to tunnel any jsg::JsExceptionThrown instances back. |
| 930 | v8::TryCatch tryCatch(js.v8Isolate); |
| 931 | ExceptionOrDuration limitErrorOrTime = 0 * kj::NANOSECONDS; |
| 932 | try { |
| 933 | auto limitScope = worker->getIsolate().getLimitEnforcer().enterDynamicImportJs( |
| 934 | lock, limitErrorOrTime); |
| 935 | return DynamicImportResult(handler()); |
| 936 | } catch (jsg::JsExceptionThrown&) { |
| 937 | // Handled below... |
| 938 | } catch (kj::Exception& ex) { |
| 939 | kj::throwFatalException(kj::mv(ex)); |
| 940 | } |
| 941 | |
| 942 | KJ_ASSERT(tryCatch.HasCaught()); |
| 943 | if (!tryCatch.CanContinue() || tryCatch.Exception().IsEmpty()) { |
| 944 | // There's nothing else we can do here but throw a generic fatal exception. |
| 945 | KJ_SWITCH_ONEOF(limitErrorOrTime) { |
| 946 | KJ_CASE_ONEOF(limitError, kj::Exception) { |
| 947 | kj::throwFatalException(kj::mv(limitError)); |
| 948 | } |
| 949 | KJ_CASE_ONEOF_DEFAULT { |
| 950 | kj::throwFatalException( |
| 951 | JSG_KJ_EXCEPTION(FAILED, Error, "Failed to load dynamic module.")); |
| 952 | } |
| 953 | } |
| 954 | } |
| 955 | return DynamicImportResult(js.v8Ref(tryCatch.Exception()), true); |
| 956 | }); |
| 957 | }); |
| 958 | }; |
| 959 | |
| 960 | modules.setDynamicImportCallback([](jsg::Lock& js, DynamicImportHandler handler) mutable { |
| 961 | KJ_IF_SOME(context, IoContext::tryCurrent()) { |
| 962 | // If we are within the scope of a IoContext, then we are going to pop |
| 963 | // out of it to perform the actual module instantiation. |
| 964 | |
| 965 | return context.awaitIo(js, |
| 966 | handleDynamicImport(kj::atomicAddRef(context.getWorker()), kj::mv(handler), |
| 967 | jsg::AsyncContextFrame::currentRef(js)), |
| 968 | [](jsg::Lock& js, DynamicImportResult result) { |
| 969 | if (result.isException) { |
| 970 | return js.rejectedPromise<jsg::Value>(kj::mv(result.value)); |
| 971 | } |
| 972 | return js.resolvedPromise(kj::mv(result.value)); |
| 973 | }); |
| 974 | } |
| 975 | |
| 976 | // If we got here, there is no current IoContext. We're going to perform the |
| 977 | // module resolution synchronously and we do not have to worry about blocking any |
| 978 | // i/o. We get here, for instance, when dynamic import is used at the top level of |
| 979 | // a script (which is weird, but allowed). |
| 980 | // |
| 981 | // We do not need to use limitEnforcer.enterDynamicImportJs() here because this should |
| 982 | // already be covered by the startup resource limiter. |
| 983 | return js.resolvedPromise(handler()); |
| 984 | }); |
| 985 | } |
| 986 | |
| 987 | kj::Maybe<const workerd::jsg::modules::ModuleRegistry&> getNewModuleRegistry() const { |
| 988 | return maybeNewModuleRegistry.map( |
| 989 | [](auto& r) -> const workerd::jsg::modules::ModuleRegistry& { return *r.get(); }); |
| 990 | } |
| 991 | }; |
| 992 | |
| 993 | namespace { |
| 994 | |
| 995 | // Given an array of strings, return a valid serialized JSON string like: |
| 996 | // {"flags":["minimal_subrequests",...]} |
| 997 | // |
| 998 | // Return null if the array is empty. |
| 999 | kj::Maybe<kj::String> makeCompatJson(kj::ArrayPtr<kj::StringPtr> enableFlags) { |
| 1000 | if (enableFlags.size() == 0) { |
| 1001 | return kj::none; |
| 1002 | } |
| 1003 | |
| 1004 | // Calculate the size of the string we're going to generate. |
| 1005 | constexpr auto PREFIX = "{\"flags\":["_kj; |
| 1006 | constexpr auto SUFFIX = "]}"_kj; |
| 1007 | uint size = std::accumulate(enableFlags.begin(), enableFlags.end(), |
| 1008 | // We need two quotes and one comma for each enable-flag past the first, plus a NUL char. |
| 1009 | PREFIX.size() + SUFFIX.size() + 3 * enableFlags.size(), |
| 1010 | [](uint z, kj::StringPtr s) { return z + s.size(); }); |
| 1011 | |
| 1012 | kj::Vector<char> json(size); |
| 1013 | |
| 1014 | json.addAll(PREFIX); |
| 1015 | |
| 1016 | bool first = true; |
| 1017 | for (auto flag: enableFlags) { |
| 1018 | if (first) { |
| 1019 | first = false; |
| 1020 | } else { |
| 1021 | json.add(','); |
| 1022 | } |
| 1023 | |
| 1024 | json.add('"'); |
| 1025 | |
| 1026 | for (auto& c: flag.asArray()) { |
| 1027 | // TODO(cleanup): Copied from simpleJsonStringCheck(). Hopefully this will |
| 1028 | // go away forever soon. |
| 1029 | KJ_REQUIRE(c != '\"'); |
| 1030 | KJ_REQUIRE(c != '\\'); |
| 1031 | KJ_REQUIRE(c >= 0x20); |
| 1032 | } |
| 1033 | json.addAll(flag); |
| 1034 | |
| 1035 | json.add('"'); |
| 1036 | } |
| 1037 | |
| 1038 | json.addAll(SUFFIX); |
| 1039 | json.add('\0'); |
| 1040 | |
| 1041 | return kj::String(json.releaseAsArray()); |
| 1042 | } |
| 1043 | |
| 1044 | // When a promise is created in a different IoContext, we need to use a |
| 1045 | // kj::CrossThreadFulfiller in order to wait on it. The Waiter instance will |
| 1046 | // be held on the Promise itself, and will be fulfilled/rejected when the |
| 1047 | // promise is resolved or rejected. This will signal all of the waiters |
| 1048 | // from other IoContexts. |
| 1049 | jsg::Promise<void> addCrossThreadPromiseWaiter(jsg::Lock& js, v8::Local<v8::Promise>& promise) { |
| 1050 | auto waiter = kj::newPromiseAndCrossThreadFulfiller<void>(); |
| 1051 | |
| 1052 | struct Waiter: public kj::Refcounted { |
| 1053 | kj::Maybe<kj::Own<kj::CrossThreadPromiseFulfiller<void>>> fulfiller; |
| 1054 | void done() { |
| 1055 | KJ_IF_SOME(f, fulfiller) { |
| 1056 | // Done this way so that the fulfiller is released as soon as possible |
| 1057 | // when done as the JS promise may not clean up reactions right away. |
| 1058 | f->fulfill(); |
| 1059 | fulfiller = kj::none; |
| 1060 | } |
| 1061 | } |
| 1062 | Waiter(kj::Own<kj::CrossThreadPromiseFulfiller<void>> fulfiller) |
| 1063 | : fulfiller(kj::mv(fulfiller)) {} |
| 1064 | }; |
| 1065 | |
| 1066 | auto fulfiller = kj::refcounted<Waiter>(kj::mv(waiter.fulfiller)); |
| 1067 | |
| 1068 | auto onSuccess = [waiter = kj::addRef(*fulfiller)]( |
| 1069 | jsg::Lock& js, jsg::Value value) mutable { waiter->done(); }; |
| 1070 | |
| 1071 | auto onFailure = [waiter = kj::mv(fulfiller)]( |
| 1072 | jsg::Lock& js, jsg::Value exception) mutable { waiter->done(); }; |
| 1073 | |
| 1074 | js.toPromise(promise).then(js, kj::mv(onSuccess), kj::mv(onFailure)); |
| 1075 | |
| 1076 | return IoContext::current().awaitIo(js, kj::mv(waiter.promise)); |
| 1077 | } |
| 1078 | |
| 1079 | struct HeapSnapshotDeleter: public kj::Disposer { |
| 1080 | static const HeapSnapshotDeleter INSTANCE; |
| 1081 | void disposeImpl(void* ptr) const override { |
| 1082 | auto snapshot = const_cast<v8::HeapSnapshot*>(static_cast<const v8::HeapSnapshot*>(ptr)); |
| 1083 | snapshot->Delete(); |
| 1084 | } |
| 1085 | }; |
| 1086 | const HeapSnapshotDeleter HeapSnapshotDeleter::INSTANCE; |
| 1087 | |
| 1088 | } // namespace |
| 1089 | |
| 1090 | Worker::Isolate::Isolate(kj::Own<Api> apiParam, |
| 1091 | kj::Own<IsolateObserver> metricsParam, |
| 1092 | kj::StringPtr id, |
| 1093 | kj::Own<IsolateLimitEnforcer> limitEnforcerParam, |
| 1094 | InspectorPolicy inspectorPolicy, |
| 1095 | LoggingOptions loggingOptions) |
| 1096 | : metrics(kj::mv(metricsParam)), |
| 1097 | id(kj::str(id)), |
| 1098 | limitEnforcer(kj::mv(limitEnforcerParam)), |
| 1099 | cpuLimitNearlyExceededCallback( |
| 1100 | kj::MutexGuarded<kj::Maybe<kj::Function<void(void)>>>(kj::none)), |
| 1101 | api(kj::mv(apiParam)), |
| 1102 | loggingOptions(loggingOptions), |
| 1103 | featureFlagsForFl(makeCompatJson(decompileCompatibilityFlagsForFl(api->getFeatureFlags()))), |
| 1104 | impl(kj::heap<Impl>(*api, *metrics, *limitEnforcer, inspectorPolicy)), |
| 1105 | weakIsolateRef(WeakIsolateRef::wrap(this)), |
| 1106 | traceAsyncContextKey(kj::refcounted<jsg::AsyncContextFrame::StorageKey>()), |
| 1107 | userTraceAsyncContextKey(kj::refcounted<jsg::AsyncContextFrame::StorageKey>()) { |
| 1108 | api->setIsolateObserver(*metrics); |
| 1109 | metrics->created(); |
| 1110 | // We just created our isolate, so we don't need to use Isolate::Impl::Lock (nor an async lock). |
| 1111 | jsg::runInV8Stack([&](jsg::V8StackScope& stackScope) { |
| 1112 | auto lock = api->lock(stackScope); |
| 1113 | auto features = api->getFeatureFlags(); |
| 1114 | |
| 1115 | KJ_DASSERT(lock->v8Isolate->GetNumberOfDataSlots() >= jsg::SET_DATA_SLOTS_IN_USE); |
| 1116 | KJ_DASSERT(lock->v8Isolate->GetData(jsg::SET_DATA_ISOLATE) == nullptr); |
| 1117 | lock->v8Isolate->SetData(jsg::SET_DATA_ISOLATE, this); |
| 1118 | |
| 1119 | lock->setCaptureThrowsAsRejections(features.getCaptureThrowsAsRejections()); |
| 1120 | // TODO(cleanup): Now that this list has grown significantly, we should probably |
| 1121 | // refactor to pass all of the options in a single call instead of one by one. |
| 1122 | if (features.getSetToStringTag()) { |
| 1123 | lock->setToStringTag(); |
| 1124 | } |
| 1125 | if (features.getShouldSetImmutablePrototype() || features.getPythonWorkers()) { |
| 1126 | lock->setImmutablePrototype(); |
| 1127 | } |
| 1128 | if (features.getSpecCompliantPropertyAttributes()) { |
| 1129 | lock->setSpecCompliantPropertyAttributes(); |
| 1130 | } |
| 1131 | if (features.getNodeJsCompatV2()) { |
| 1132 | lock->setNodeJsCompatEnabled(); |
| 1133 | } |
| 1134 | if (features.getEnableNodeJsProcessV2()) { |
| 1135 | lock->setNodeJsProcessV2Enabled(); |
| 1136 | } |
| 1137 | if (features.getRequireReturnsDefaultExport()) { |
| 1138 | lock->setRequireReturnsDefaultExportEnabled(); |
| 1139 | } |
| 1140 | if (features.getThrowOnUnrecognizedImportAssertion()) { |
| 1141 | lock->setThrowOnUnrecognizedImportAssertion(); |
| 1142 | } |
| 1143 | if (features.getNoTopLevelAwaitInRequire()) { |
| 1144 | lock->disableTopLevelAwait(); |
| 1145 | } |
| 1146 | if (features.getEnhancedErrorSerialization()) { |
| 1147 | lock->setUsingEnhancedErrorSerialization(); |
| 1148 | } |
| 1149 | if (features.getFastJsgStruct()) { |
| 1150 | lock->setUsingFastJsgStruct(); |
| 1151 | } |
| 1152 | |
| 1153 | if (impl->inspector != kj::none || ::kj::_::Debug::shouldLog(::kj::LogSeverity::INFO)) { |
| 1154 | lock->setLoggerCallback([this](jsg::Lock& js, kj::StringPtr message) { |
| 1155 | if (impl->inspector != kj::none) { |
| 1156 | logMessage(js, static_cast<uint16_t>(cdp::LogType::WARNING), message); |
| 1157 | } |
| 1158 | KJ_LOG(INFO, "console warning", message); |
| 1159 | }); |
| 1160 | lock->setErrorReporterCallback([this](jsg::Lock& js, kj::String desc, |
| 1161 | const jsg::JsValue& error, const jsg::JsMessage& message) { |
| 1162 | // Only add exception to trace when running within an I/O context with a tracer. |
| 1163 | KJ_IF_SOME(ioContext, IoContext::tryCurrent()) { |
| 1164 | KJ_IF_SOME(tracer, ioContext.getWorkerTracer()) { |
| 1165 | addExceptionToTrace(js, ioContext, tracer, UncaughtExceptionSource::REQUEST_HANDLER, |
| 1166 | error, api->getErrorInterfaceTypeHandler(js)); |
| 1167 | } |
| 1168 | } |
| 1169 | |
| 1170 | KJ_IF_SOME(i, impl->inspector) { |
| 1171 | jsg::sendExceptionToInspector(js, *i.get(), kj::str(desc), error, message); |
| 1172 | } |
| 1173 | |
| 1174 | // Run with --verbose to log JS exceptions to stderr. Useful when running tests. |
| 1175 | KJ_LOG(INFO, "uncaught exception", desc); |
| 1176 | }); |
| 1177 | } |
| 1178 | |
| 1179 | // By default, V8's memory pressure level is "none". This tells V8 that no one else on the |
| 1180 | // machine is competing for memory so it might as well use all it wants and be lazy about GC. |
| 1181 | // |
| 1182 | // In our production environment, however, we can safely assume that there is always memory |
| 1183 | // pressure, because every machine is handling thousands of tenants all the time. So we might |
| 1184 | // as well just throw the switch to "moderate" right away. |
| 1185 | lock->v8Isolate->MemoryPressureNotification(v8::MemoryPressureLevel::kModerate); |
| 1186 | |
| 1187 | // Register GC prologue and epilogue callbacks so that we can report GC CPU time via the |
| 1188 | // "request_context" Jaeger span. |
| 1189 | lock->v8Isolate->AddGCPrologueCallback( |
| 1190 | [](v8::Isolate* isolate, v8::GCType type, v8::GCCallbackFlags flags, void* data) noexcept { |
| 1191 | // We assume that a v8::Locker is alive during GC. |
| 1192 | KJ_DASSERT(v8::Locker::IsLocked(isolate)); |
| 1193 | auto& self = *reinterpret_cast<Isolate*>(data); |
| 1194 | // However, currentLock might not be available, if (like in our Worker::Isolate constructor) we |
| 1195 | // don't use a Worker::Isolate::Impl::Lock. |
| 1196 | KJ_IF_SOME(currentLock, self.impl->currentLock) { |
| 1197 | currentLock.gcPrologue(); |
| 1198 | } |
| 1199 | }, this); |
| 1200 | lock->v8Isolate->AddGCEpilogueCallback( |
| 1201 | [](v8::Isolate* isolate, v8::GCType type, v8::GCCallbackFlags flags, void* data) noexcept { |
| 1202 | // We make similar assumptions about v8::Locker and currentLock as in the prologue callback. |
| 1203 | KJ_DASSERT(v8::Locker::IsLocked(isolate)); |
| 1204 | auto& self = *reinterpret_cast<Isolate*>(data); |
| 1205 | KJ_IF_SOME(currentLock, self.impl->currentLock) { |
| 1206 | currentLock.gcEpilogue(); |
| 1207 | } |
| 1208 | }, this); |
| 1209 | lock->v8Isolate->SetPromiseRejectCallback([](v8::PromiseRejectMessage message) { |
| 1210 | // TODO(cleanup): IoContext doesn't really need to be involved here. We are trying to call |
| 1211 | // a method of ServiceWorkerGlobalScope, which is the context object. So we should be able to |
| 1212 | // do something like unwrap(lock, isolate->GetCurrentContext()).emitPromiseRejection(). |
| 1213 | // However, JSG doesn't currently provide an easy way to do this. |
| 1214 | KJ_IF_SOME(ioContext, IoContext::tryCurrent()) { |
| 1215 | try { |
| 1216 | ioContext.getCurrentLock().reportPromiseRejectEvent(message); |
| 1217 | } catch (jsg::JsExceptionThrown&) { |
| 1218 | // V8 expects us to just return. |
| 1219 | return; |
| 1220 | } |
| 1221 | } |
| 1222 | }); |
| 1223 | |
| 1224 | // The PromiseCrossContextCallback is used to allow cross-IoContext promise following. |
| 1225 | // When the IoContext scope is entered, we set the "promise context tag" associated |
| 1226 | // with the IoContext on the Isolate that is locked. Any Promise that is created within |
| 1227 | // that scope will be tagged with the same promise context tag. When an attempt to |
| 1228 | // follow a promise occurs (e.g. either using Promise.prototype.then() or await, etc) |
| 1229 | // our patched v8 logic will check to see if the followed promise's tag matches the |
| 1230 | // current Isolate tag. If they do not, then v8 will invoke this callback. The promise |
| 1231 | // here is the promise that belongs to a different IoContext. |
| 1232 | lock->v8Isolate->SetPromiseCrossContextCallback( |
| 1233 | [](v8::Local<v8::Context> context, v8::Local<v8::Promise> promise, |
| 1234 | v8::Local<v8::Object> tag) -> v8::MaybeLocal<v8::Promise> { |
| 1235 | auto& js = jsg::Lock::current(); |
| 1236 | try { |
| 1237 | // Generally this condition is only going to happen when using dynamic imports. |
| 1238 | // It should not be common. |
| 1239 | JSG_REQUIRE(IoContext::hasCurrent(), Error, |
| 1240 | "Unable to wait on a promise created within a request when not running within a " |
| 1241 | "request."); |
| 1242 | |
| 1243 | return js.wrapSimplePromise( |
| 1244 | addCrossThreadPromiseWaiter(js, promise) |
| 1245 | .then(js, [promise = js.v8Ref(promise.As<v8::Value>())](auto& js) mutable { |
| 1246 | // Once the waiter has been resolved, return the now settled promise. |
| 1247 | // Since the promise has been settled, it is now safe to access from |
| 1248 | // other requests. Note that the resolved value of the promise still |
| 1249 | // might not be safe to access! (e.g. if it contains any IoOwns attached |
| 1250 | // to the other request IoContext). |
| 1251 | return kj::mv(promise); |
| 1252 | })); |
| 1253 | } catch (jsg::JsExceptionThrown&) { |
| 1254 | // Exceptions here are generally unexpected but possible because the jsg::Promise |
| 1255 | // then can fail if the isolate is in the process of being torn down. Let's just |
| 1256 | // return control back to V8 which should handle the case. |
| 1257 | return v8::MaybeLocal<v8::Promise>(); |
| 1258 | } catch (...) { |
| 1259 | auto ex = kj::getCaughtExceptionAsKj(); |
| 1260 | KJ_LOG(ERROR, "Setting promise cross context follower failed unexpectedly", ex); |
| 1261 | jsg::throwInternalError(js.v8Isolate, kj::mv(ex)); |
| 1262 | return v8::MaybeLocal<v8::Promise>(); |
| 1263 | } |
| 1264 | }); |
| 1265 | |
| 1266 | // The PromiseCrossContextResolveCallback is used to ensure that promise reactions |
| 1267 | // are only scheduled on the microtask queue from the appropriate IoContext for the |
| 1268 | // promise. Huh? Yeah, that's not super clear... let me explain a bit more. |
| 1269 | // Every request runs in its own IoContext. |
| 1270 | // Some I/O objects are bound to the IoContext when they are created. |
| 1271 | // If these objects are accessed from the wrong IoContext, things blow up. |
| 1272 | // If I create a promise in one request and pass the resolve/reject functions |
| 1273 | // off to a different request, bad things can happen because the IoContext can |
| 1274 | // actually change in the promise continuation. Take the following case for example: |
| 1275 | // |
| 1276 | // In request one: |
| 1277 | // |
| 1278 | // const ab = AbortSignal.abort(); // AbortSignal is bound to the IoContext |
| 1279 | // const { promise, resolve } = Promise.withResolvers(); |
| 1280 | // globalThis.resolve = resolve; |
| 1281 | // await promise; |
| 1282 | // console.log(ab.aborted); |
| 1283 | // |
| 1284 | // In request two: |
| 1285 | // |
| 1286 | // globalThis.resolve(); |
| 1287 | // |
| 1288 | // What previously would happen is that the `console.log(ab.aborted) after the |
| 1289 | // `await promise` in request one would fail with an error because the current |
| 1290 | // IoContext would change! (it would be the IoContext from request two!). |
| 1291 | // |
| 1292 | // That's bad. |
| 1293 | // |
| 1294 | // So this callback is added to ensure that the promise reactions for the promise |
| 1295 | // being resolved are not scheduled until we are back in the correct IoContext for |
| 1296 | // the promise. |
| 1297 | // |
| 1298 | // This happens by (ab)using the DeleteQueue that is specific to the owning |
| 1299 | // IoContext. When the IoContext is entered, the isolate is updated with a |
| 1300 | // current "promise tag". Whenever a promise is created, it is associated with |
| 1301 | // the isolate's current tag. Whenever a promise is followed (calling .then, etc), |
| 1302 | // we check the tag and arrange for a cross-thread resolve. When the promise is |
| 1303 | // resolved or rejected, we check the tag also. If the promise tag and the current |
| 1304 | // isolate tag do not match, the function below is called. |
| 1305 | if (features.getHandleCrossRequestPromiseResolution()) { |
| 1306 | lock->v8Isolate->SetPromiseCrossContextResolveCallback( |
| 1307 | [](v8::Isolate* isolate, v8::Local<v8::Value> tag, v8::Local<v8::Data> reactions, |
| 1308 | v8::Local<v8::Value> argument, |
| 1309 | std::function<void(v8::Isolate * isolate, v8::Local<v8::Data> reactions, |
| 1310 | v8::Local<v8::Value> argument)> callback) -> v8::Maybe<void> { |
| 1311 | try { |
| 1312 | auto& js = jsg::Lock::from(isolate); |
| 1313 | |
| 1314 | // The promise tag is generally opaque except for right here. The tag |
| 1315 | // wraps an instanceof kj::Own<IoCrossContextExecutor>, which wraps an atomically |
| 1316 | // refcounted pointer to the DeleteQueue for the correct isolate. |
| 1317 | // We simply pass the given callback, reactions, and argument to |
| 1318 | // a function that will be added to the queue inside DeleteQueue. |
| 1319 | // The next time the relevant IoContext is entered, this queue will |
| 1320 | // be drained and the actions will be run. Adding the task to the |
| 1321 | // delete queue will also signal the IoContext that it should wake |
| 1322 | // up and drain the queue. Simple, eh? |
| 1323 | // |
| 1324 | // A word of warning tho! It is possible for the IoContext to be |
| 1325 | // destroyed before the promise is resolved. Any actions that have |
| 1326 | // already been added to the queue would end up being dropped silently |
| 1327 | // on the floor. Actions that are added to the queue now will be run |
| 1328 | // immediately in the wrong IoContext. |
| 1329 | auto& ref = jsg::unwrapOpaqueRef<kj::Own<IoCrossContextExecutor>>(isolate, tag); |
| 1330 | ref->execute(js, |
| 1331 | [reactions = jsg::Data(isolate, reactions), argument = jsg::V8Ref(isolate, argument), |
| 1332 | callback = kj::mv(callback)](jsg::Lock& js) mutable { |
| 1333 | callback(js.v8Isolate, reactions.getHandle(js), argument.getHandle(js)); |
| 1334 | }); |
| 1335 | return v8::JustVoid(); |
| 1336 | } catch (jsg::JsExceptionThrown&) { |
| 1337 | // Exceptions here are generally unexpected but possible because the jsg::Promise |
| 1338 | // then can fail if the isolate is in the process of being torn down. Let's just |
| 1339 | // return control back to V8 which should handle the case. |
| 1340 | // Note that errors thrown here and below should cause the resolve() or reject() |
| 1341 | // function calls to throw, which is unusual. Just important to keep that in mind. |
| 1342 | // Most likely errors thrown here are fatal so that should be OK. |
| 1343 | return v8::Nothing<void>(); |
| 1344 | } catch (...) { |
| 1345 | jsg::throwInternalError(isolate, kj::getCaughtExceptionAsKj()); |
| 1346 | return v8::Nothing<void>(); |
| 1347 | } |
| 1348 | }); |
| 1349 | } |
| 1350 | }); |
| 1351 | } |
| 1352 | |
| 1353 | Worker::Script::Script(kj::Own<const Isolate> isolateParam, |
| 1354 | kj::StringPtr id, |
| 1355 | const Script::Source& source, |
| 1356 | IsolateObserver::StartType startType, |
| 1357 | bool logNewScript, |
| 1358 | kj::Maybe<ValidationErrorReporter&> errorReporter, |
| 1359 | kj::Maybe<kj::Own<api::pyodide::ArtifactBundler_State>> artifacts, |
| 1360 | SpanParent parentSpan, |
| 1361 | kj::Own<workerd::VirtualFileSystem> vfs, |
| 1362 | kj::Maybe<kj::Arc<workerd::jsg::modules::ModuleRegistry>> maybeNewModuleRegistry) |
| 1363 | : isolate(kj::mv(isolateParam)), |
| 1364 | id(kj::str(id)), |
| 1365 | modular(source.variant.is<ModulesSource>()), |
| 1366 | python(modular && source.variant.get<ModulesSource>().isPython), |
| 1367 | impl(kj::heap<Impl>(kj::mv(vfs), kj::mv(maybeNewModuleRegistry))), |
| 1368 | dynamicEnvBuilder(source.dynamicEnvBuilder.map( |
| 1369 | [](const auto& inst) -> kj::Arc<DynamicEnvBuilder> { return inst.addRef(); })) { |
| 1370 | auto parseMetrics = isolate->metrics->parse(startType); |
| 1371 | // TODO(perf): It could make sense to take an async lock when constructing a script if we |
| 1372 | // co-locate multiple scripts in the same isolate. As of this writing, we do not, except in |
| 1373 | // previews, where it doesn't matter. If we ever do co-locate multiple scripts in the same |
| 1374 | // isolate, we may wish to make the RequestObserver object available here, in order to |
| 1375 | // attribute lock timing to that request. |
| 1376 | jsg::runInV8Stack([&](jsg::V8StackScope& stackScope) { |
| 1377 | Isolate::Impl::Lock recordedLock( |
| 1378 | *isolate, Worker::Lock::TakeSynchronously(kj::none), stackScope); |
| 1379 | auto& lock = *recordedLock.lock; |
| 1380 | |
| 1381 | // If we throw an exception, it's important that `impl` is destroyed under lock. |
| 1382 | KJ_ON_SCOPE_FAILURE({ |
| 1383 | auto implToDestroy = kj::mv(impl); |
| 1384 | KJ_IF_SOME(c, implToDestroy->moduleContext) { |
| 1385 | recordedLock.disposeContext(kj::mv(c)); |
| 1386 | } else { |
| 1387 | // Else block to avoid dangling else clang warning. |
| 1388 | } |
| 1389 | }); |
| 1390 | |
| 1391 | lock.withinHandleScope([&] { |
| 1392 | if (isolate->impl->inspector != kj::none || errorReporter != kj::none) { |
| 1393 | lock.v8Isolate->SetCaptureStackTraceForUncaughtExceptions(true); |
| 1394 | } |
| 1395 | |
| 1396 | v8::Local<v8::Context> context; |
| 1397 | if (modular) { |
| 1398 | // Modules can't be compiled for multiple contexts. We need to create the real context now. |
| 1399 | auto& mContext = impl->moduleContext.emplace(isolate->getApi().newContext(lock, |
| 1400 | { |
| 1401 | .newModuleRegistry = impl->getNewModuleRegistry(), |
| 1402 | .schemaLoader = getSchemaLoader(), |
| 1403 | })); |
| 1404 | mContext->enableWarningOnSpecialEvents(); |
| 1405 | context = mContext.getHandle(lock); |
| 1406 | recordedLock.setupContext(context); |
| 1407 | } else { |
| 1408 | // Although we're going to compile a script independent of context, V8 requires that |
| 1409 | // there be an active context, otherwise it will segfault, I guess. So we create a |
| 1410 | // dummy context. (Undocumented, as usual.) |
| 1411 | context = |
| 1412 | v8::Context::New(lock.v8Isolate, nullptr, v8::ObjectTemplate::New(lock.v8Isolate)); |
| 1413 | // We need to set the highest used index in every context we create to be a nullptr |
| 1414 | // This is because we might later on call GetAlignedPointerFromEmbedderData which fails with |
| 1415 | // a fatal error if the array is smaller than the given index. |
| 1416 | jsg::setAlignedPointerInEmbedderData( |
| 1417 | context, jsg::ContextPointerSlot::MAX_POINTER_SLOT, nullptr); |
| 1418 | } |
| 1419 | |
| 1420 | JSG_WITHIN_CONTEXT_SCOPE(lock, context, [&](jsg::Lock& js) { |
| 1421 | // const_cast OK because we hold the isolate lock. |
| 1422 | Worker::Isolate& lockedWorkerIsolate = const_cast<Isolate&>(*isolate); |
| 1423 | |
| 1424 | if (logNewScript) { |
| 1425 | // HACK: Log a message indicating that a new script was loaded. This is used only when the |
| 1426 | // inspector is enabled. We want to do this immediately after the context is created, |
| 1427 | // before the user gets a chance to modify the behavior of the console, which if they |
| 1428 | // did, we'd then need to be more careful to apply time limits and such. |
| 1429 | lockedWorkerIsolate.logMessage(lock, static_cast<uint16_t>(cdp::LogType::WARNING), |
| 1430 | "Script modified; context reset."); |
| 1431 | } |
| 1432 | |
| 1433 | // We need to register this context with the inspector, otherwise errors won't be |
| 1434 | // reported. But we want it to be un-registered as soon as the script has been |
| 1435 | // compiled, otherwise the inspector will end up with multiple contexts active which |
| 1436 | // is very confusing for the user (since they'll have to select from the drop-down |
| 1437 | // which context to use). |
| 1438 | // |
| 1439 | // (For modules, the context was already registered by `setupContext()`, above. |
| 1440 | KJ_IF_SOME(i, isolate->impl->inspector) { |
| 1441 | if (!modular) { |
| 1442 | i.get()->contextCreated( |
| 1443 | v8_inspector::V8ContextInfo(context, 1, jsg::toInspectorStringView("Compiler"))); |
| 1444 | } |
| 1445 | } else { |
| 1446 | } // Here to squash a compiler warning |
| 1447 | KJ_DEFER({ |
| 1448 | if (!modular) { |
| 1449 | KJ_IF_SOME(i, isolate->impl->inspector) { |
| 1450 | i.get()->contextDestroyed(context); |
| 1451 | } else { |
| 1452 | } // Here to squash a compiler warning |
| 1453 | } |
| 1454 | }); |
| 1455 | |
| 1456 | v8::TryCatch catcher(lock.v8Isolate); |
| 1457 | ExceptionOrDuration limitErrorOrTime = 0 * kj::NANOSECONDS; |
| 1458 | |
| 1459 | try { |
| 1460 | try { |
| 1461 | KJ_SWITCH_ONEOF(source.variant) { |
| 1462 | KJ_CASE_ONEOF(script, ScriptSource) { |
| 1463 | // This path is used for the older, service worker syntax workers. |
| 1464 | |
| 1465 | if (script.capnpSchemas.size() > 0) { |
| 1466 | // const_cast OK because we hold the isolate lock. |
| 1467 | auto& schemaLoader = const_cast<capnp::SchemaLoader&>(getSchemaLoader()); |
| 1468 | for (auto node: script.capnpSchemas) { |
| 1469 | schemaLoader.load(node); |
| 1470 | } |
| 1471 | } |
| 1472 | |
| 1473 | impl->globals = |
| 1474 | isolate->getApi().compileServiceWorkerGlobals(lock, script, *isolate); |
| 1475 | |
| 1476 | { |
| 1477 | // It's unclear to me if CompileUnboundScript() can get trapped in any |
| 1478 | // infinite loops or excessively-expensive computation requiring a time |
| 1479 | // limit. We'll go ahead and apply a time limit just to be safe. Don't |
| 1480 | // add it to the rollover bank, though. |
| 1481 | auto limitScope = |
| 1482 | isolate->getLimitEnforcer().enterStartupJs(lock, limitErrorOrTime); |
| 1483 | impl->unboundScriptOrMainModule = |
| 1484 | jsg::NonModuleScript::compile(lock, script.mainScript, script.mainScriptName); |
| 1485 | } |
| 1486 | } |
| 1487 | |
| 1488 | KJ_CASE_ONEOF(modulesSource, ModulesSource) { |
| 1489 | // This path is used for the new ESM worker syntax. |
| 1490 | |
| 1491 | if (modulesSource.capnpSchemas.size() > 0) { |
| 1492 | // const_cast OK because we hold the isolate lock. |
| 1493 | auto& schemaLoader = const_cast<capnp::SchemaLoader&>(getSchemaLoader()); |
| 1494 | for (auto node: modulesSource.capnpSchemas) { |
| 1495 | schemaLoader.load(node); |
| 1496 | } |
| 1497 | } |
| 1498 | |
| 1499 | if (!isolate->getApi().getFeatureFlags().getNewModuleRegistry()) { |
| 1500 | kj::Own<void> limitScope; |
| 1501 | if (modulesSource.isPython) { |
| 1502 | limitScope = |
| 1503 | isolate->getLimitEnforcer().enterStartupPython(js, limitErrorOrTime); |
| 1504 | } else { |
| 1505 | limitScope = isolate->getLimitEnforcer().enterStartupJs(js, limitErrorOrTime); |
| 1506 | } |
| 1507 | impl->configureDynamicImports(lock, *jsg::ModuleRegistry::from(lock)); |
| 1508 | isolate->getApi().compileModules( |
| 1509 | lock, modulesSource, *isolate, kj::mv(artifacts), parentSpan.addRef()); |
| 1510 | } |
| 1511 | impl->unboundScriptOrMainModule = kj::Path::parse(modulesSource.mainModule); |
| 1512 | } |
| 1513 | } |
| 1514 | |
| 1515 | parseMetrics->done(); |
| 1516 | } catch (const kj::Exception& e) { |
| 1517 | lock.throwException(e.clone()); |
| 1518 | // lock.throwException() here will throw a jsg::JsExceptionThrown which we catch |
| 1519 | // in the outer try/catch. |
| 1520 | } |
| 1521 | } catch (const jsg::JsExceptionThrown&) { |
| 1522 | reportStartupError(id, lock, isolate->impl->inspector, isolate->getLimitEnforcer(), |
| 1523 | kj::mv(limitErrorOrTime), catcher, errorReporter, impl->permanentException, |
| 1524 | parentSpan.addRef(), dynamicEnvBuilder != kj::none); |
| 1525 | } |
| 1526 | }); |
| 1527 | }); |
| 1528 | }); |
| 1529 | } |
| 1530 | |
| 1531 | void Worker::Script::installVirtualFileSystemOnContext(v8::Local<v8::Context> context) const { |
| 1532 | jsg::setAlignedPointerInEmbedderData(context, jsg::ContextPointerSlot::VIRTUAL_FILE_SYSTEM, |
| 1533 | const_cast<VirtualFileSystem*>(impl->vfs.get())); |
| 1534 | } |
| 1535 | |
| 1536 | const capnp::SchemaLoader& Worker::Script::getSchemaLoader() const { |
| 1537 | KJ_IF_SOME(moduleRegistry, impl->maybeNewModuleRegistry) { |
| 1538 | return moduleRegistry->getSchemaLoader(); |
| 1539 | } else { |
| 1540 | return *KJ_ASSERT_NONNULL(impl->maybeSchemaLoader); |
| 1541 | } |
| 1542 | } |
| 1543 | |
| 1544 | kj::Own<const Worker::Isolate::WeakIsolateRef> Worker::Isolate::getWeakRef() const { |
| 1545 | return weakIsolateRef->addRef(); |
| 1546 | } |
| 1547 | |
| 1548 | kj::StringPtr Worker::Isolate::getUuid() const { |
| 1549 | // As of this writing, getUuid() is only used by actors, for metrics. We don't want to bother |
| 1550 | // generating it if not used. The call site does not have nor want an isolate lock, so we use a |
| 1551 | // kj::Lazy to make initialization thread-safe. |
| 1552 | return impl->uuid.get( |
| 1553 | [](kj::SpaceFor<kj::String>& space) { return space.construct(randomUUID(kj::none)); }); |
| 1554 | } |
| 1555 | |
| 1556 | Worker::Isolate::~Isolate() noexcept(false) { |
| 1557 | metrics->teardownStarted(); |
| 1558 | |
| 1559 | // Update the isolate stats one last time to make sure we're accurate for cleanup in |
| 1560 | // `evicted()`. |
| 1561 | limitEnforcer->reportMetrics(*metrics); |
| 1562 | |
| 1563 | metrics->evicted(); |
| 1564 | weakIsolateRef->invalidate(); |
| 1565 | // The cpuLimitNearlyExceededCallback may hold references to objects owned by the isolate and |
| 1566 | // their destructors need the isolate to still exist. So destroy them before we destroy the |
| 1567 | // isolate. |
| 1568 | *cpuLimitNearlyExceededCallback.lockExclusive() = kj::none; |
| 1569 | |
| 1570 | // Make sure to destroy things under lock. This lock should never be contended since the isolate |
| 1571 | // is about to be destroyed, but we have to take the lock in order to enter the isolate. |
| 1572 | // It's also important that we lock one last time, in order to destroy any remaining workers in |
| 1573 | // worker destruction queue. |
| 1574 | jsg::runInV8Stack([&](jsg::V8StackScope& stackScope) { |
| 1575 | Isolate::Impl::Lock recordedLock(*this, Worker::Lock::TakeSynchronously(kj::none), stackScope); |
| 1576 | metrics->teardownLockAcquired(); |
| 1577 | auto inspector = kj::mv(impl->inspector); |
| 1578 | auto dropTraceAsyncContextKey = kj::mv(traceAsyncContextKey); |
| 1579 | auto dropUserTraceAsyncContextKey = kj::mv(userTraceAsyncContextKey); |
| 1580 | // The Rust Realm must be dropped under lock since Realm::drop() accesses V8 globals |
| 1581 | // and calls drop functions that may interact with V8. |
| 1582 | auto dropRealm = kj::mv(impl->realm); |
| 1583 | |
| 1584 | // Release all tracked WASM instance entries while V8 is still alive. Each entry holds a |
| 1585 | // shared_ptr<v8::BackingStore> whose destructor who needs the isolate to still be alive. |
| 1586 | // This is analogous to the cpuTimeLimitNearlyExceededCallback detaching above ^^^ |
| 1587 | limitEnforcer->getTrackedWasmInstances().clear(*recordedLock.lock); |
| 1588 | }); |
| 1589 | } |
| 1590 | |
| 1591 | Worker::Script::~Script() noexcept(false) { |
| 1592 | // Make sure to destroy things under lock. |
| 1593 | // TODO(perf): It could make sense to try to obtain an async lock before destroying a script if |
| 1594 | // multiple scripts are co-located in the same isolate. As of this writing, that doesn't happen |
| 1595 | // except in preview. In any case, Scripts are destroyed in the GC thread, where we don't care |
| 1596 | // too much about lock latency. |
| 1597 | jsg::runInV8Stack([&](jsg::V8StackScope& stackScope) { |
| 1598 | Isolate::Impl::Lock recordedLock( |
| 1599 | *isolate, Worker::Lock::TakeSynchronously(kj::none), stackScope); |
| 1600 | KJ_IF_SOME(c, impl->moduleContext) { |
| 1601 | recordedLock.disposeContext(kj::mv(c)); |
| 1602 | } |
| 1603 | impl = nullptr; |
| 1604 | }); |
| 1605 | } |
| 1606 | |
| 1607 | const Worker::Isolate& Worker::Isolate::from(jsg::Lock& js) { |
| 1608 | auto ptr = js.v8Isolate->GetData(jsg::SET_DATA_ISOLATE); |
| 1609 | KJ_ASSERT(ptr != nullptr); |
| 1610 | return *static_cast<const Worker::Isolate*>(ptr); |
| 1611 | } |
| 1612 | |
| 1613 | bool Worker::Isolate::Impl::Lock::checkInWithLimitEnforcer(Worker::Isolate& isolate) { |
| 1614 | shouldReportIsolateMetrics = true; |
| 1615 | return limitEnforcer.exitJs(*lock); |
| 1616 | } |
| 1617 | |
| 1618 | kj::Maybe<kj::Function<void(void)>> Worker::Isolate::getCpuLimitNearlyExceededCallback() const { |
| 1619 | auto lock = cpuLimitNearlyExceededCallback.lockExclusive(); |
| 1620 | KJ_IF_SOME(cb, *lock) { |
| 1621 | return cb.reference(); |
| 1622 | } |
| 1623 | return kj::none; |
| 1624 | } |
| 1625 | |
| 1626 | void Worker::Isolate::setCpuLimitNearlyExceededCallback(kj::Function<void(void)> cb) const { |
| 1627 | auto lock = cpuLimitNearlyExceededCallback.lockExclusive(); |
| 1628 | // Make sure we don't reassign the callback so we don't invalidate references we've passed out. |
| 1629 | if (*lock == kj::none) { |
| 1630 | *lock = kj::mv(cb); |
| 1631 | return; |
| 1632 | } |
| 1633 | kj::throwRecoverableException(KJ_EXCEPTION( |
| 1634 | FAILED, "Python Workers Internal Error: CpuLimitNearlyExceededCallback already set")); |
| 1635 | } |
| 1636 | |
| 1637 | void Worker::Isolate::registerTrackedWasmInstance(jsg::Lock& js, |
| 1638 | v8::Local<v8::Object> instance, |
| 1639 | kj::Array<kj::byte> memory, |
| 1640 | kj::Maybe<uint32_t> signalOffset, |
| 1641 | kj::Maybe<uint32_t> terminatedOffset) const { |
| 1642 | // Register the WASM module for receiving shutdown signals. The signal handler will |
| 1643 | // iterate the list unconditionally when CPU time is nearly exhausted. |
| 1644 | KJ_IF_SOME(entry, |
| 1645 | limitEnforcer->getTrackedWasmInstances().registerSignal( |
| 1646 | js, kj::mv(memory), signalOffset, terminatedOffset)) { |
| 1647 | // Set up a weak reference to the instance. When V8 collects it, the handle becomes |
| 1648 | // empty and the GC prologue filter removes the entry, releasing the strong memory ref. |
| 1649 | entry.instanceRef.Reset(js.v8Isolate, instance); |
| 1650 | entry.instanceRef.SetWeak(); |
| 1651 | } |
| 1652 | } |
| 1653 | |
| 1654 | // EW-1319: Set WebAssembly.Module @@HasInstance |
| 1655 | // |
| 1656 | // The instanceof operator can be changed by setting the @@HasInstance method |
| 1657 | // on the object, https://tc39.es/ecma262/#sec-instanceofoperator. |
| 1658 | void setWebAssemblyModuleHasInstance(jsg::Lock& lock, v8::Local<v8::Context> context) { |
| 1659 | JSG_WITHIN_CONTEXT_SCOPE(lock, context, [&](jsg::Lock& lock) { |
| 1660 | auto instanceof = [](const v8::FunctionCallbackInfo<v8::Value>& info) { |
| 1661 | jsg::Lock::from(info.GetIsolate()).withinHandleScope([&] { |
| 1662 | info.GetReturnValue().Set(info[0]->IsWasmModuleObject()); |
| 1663 | }); |
| 1664 | }; |
| 1665 | v8::Local<v8::Function> function = jsg::check(v8::Function::New(context, instanceof)); |
| 1666 | |
| 1667 | auto webAssembly = |
| 1668 | KJ_ASSERT_NONNULL(lock.global().get(lock, "WebAssembly").tryCast<jsg::JsObject>()); |
| 1669 | auto module = KJ_ASSERT_NONNULL(webAssembly.get(lock, "Module").tryCast<jsg::JsObject>()); |
| 1670 | |
| 1671 | jsg::check(v8::Local<v8::Object>(module)->DefineOwnProperty( |
| 1672 | context, v8::Symbol::GetHasInstance(lock.v8Isolate), function)); |
| 1673 | }); |
| 1674 | } |
| 1675 | |
| 1676 | // Installs a shim around WebAssembly.instantiate and WebAssembly.Instance that hooks into the |
| 1677 | // shutdown signal if it exists |
| 1678 | void shimWebAssemblyInstantiate(jsg::Lock& lock, v8::Local<v8::Context> context) { |
| 1679 | // We need to enter the context because this function compiles and executes JavaScript via |
| 1680 | // v8::Script::Compile/Run. setupContext() is called before JSG_WITHIN_CONTEXT_SCOPE, so the |
| 1681 | // context is not yet entered at this point. |
| 1682 | v8::Context::Scope contextScope(context); |
| 1683 | |
| 1684 | // Create a C++ callback that the JS shims call to register a {instance, memory, signalOffset, |
| 1685 | // terminatedOffset} tuple. |
| 1686 | // __registerTrackedWasmInstance(instance: WebAssembly.Instance, |
| 1687 | // memory: WebAssembly.Memory, signalOffset: number, |
| 1688 | // terminatedOffset: number) |
| 1689 | // signalOffset or terminatedOffset may be -1, indicating the corresponding export is absent. |
| 1690 | auto registerCb = [](const v8::FunctionCallbackInfo<v8::Value>& info) { |
| 1691 | auto& js = jsg::Lock::from(info.GetIsolate()); |
| 1692 | js.withinHandleScope([&] { |
| 1693 | if (info.Length() < 4 || !info[0]->IsObject() || !info[1]->IsWasmMemoryObject() || |
| 1694 | !info[2]->IsNumber() || !info[3]->IsNumber()) { |
| 1695 | js.v8Isolate->ThrowException( |
| 1696 | js.str("registerTrackedWasmInstance: expected " |
| 1697 | "(WebAssembly.Instance, WebAssembly.Memory, number, number)"_kj)); |
| 1698 | return; |
| 1699 | } |
| 1700 | auto instance = info[0].As<v8::Object>(); |
| 1701 | auto memory = info[1].As<v8::WasmMemoryObject>(); |
| 1702 | // signalOffset is -1 when __instance_signal was not exported. |
| 1703 | auto signalRaw = info[2].As<v8::Number>()->Value(); |
| 1704 | kj::Maybe<uint32_t> signalOffset; |
| 1705 | if (signalRaw >= 0) { |
| 1706 | signalOffset = static_cast<uint32_t>(signalRaw); |
| 1707 | } |
| 1708 | // terminatedOffset is -1 when __instance_terminated was not exported. |
| 1709 | auto terminatedRaw = info[3].As<v8::Number>()->Value(); |
| 1710 | kj::Maybe<uint32_t> terminatedOffset; |
| 1711 | if (terminatedRaw >= 0) { |
| 1712 | terminatedOffset = static_cast<uint32_t>(terminatedRaw); |
| 1713 | } |
| 1714 | auto backingStore = memory->Buffer()->GetBackingStore(); |
| 1715 | auto wasmMemory = |
| 1716 | kj::arrayPtr(static_cast<kj::byte*>(backingStore->Data()), backingStore->ByteLength()) |
| 1717 | .attach(kj::mv(backingStore)); |
| 1718 | KJ_IF_SOME(e, kj::runCatchingExceptions([&] { |
| 1719 | Worker::Isolate::from(js).registerTrackedWasmInstance( |
| 1720 | js, instance, kj::mv(wasmMemory), signalOffset, terminatedOffset); |
| 1721 | })) { |
| 1722 | js.v8Isolate->ThrowException(js.exceptionToJs(kj::mv(e)).getHandle(js)); |
| 1723 | } |
| 1724 | }); |
| 1725 | }; |
| 1726 | auto registerFn = jsg::check(v8::Function::New(context, registerCb)); |
| 1727 | |
| 1728 | // Build the shim in JavaScript. It wraps both WebAssembly.instantiate (async) and |
| 1729 | // WebAssembly.Instance (sync constructor). |
| 1730 | auto shimScript = |
| 1731 | jsg::NonModuleScript::compile(lock, WASM_INSTANTIATE_SHIM, "wasm-instantiate-shim.js"_kj); |
| 1732 | auto shimFn = KJ_ASSERT_NONNULL(shimScript.runAndReturn(lock).tryCast<jsg::JsFunction>()); |
| 1733 | |
| 1734 | // Call the factory โ it mutates `WebAssembly` in place. |
| 1735 | shimFn.call(lock, lock.global(), jsg::JsFunction(registerFn)); |
| 1736 | } |
| 1737 | |
| 1738 | void Worker::setupContext( |
| 1739 | jsg::Lock& lock, v8::Local<v8::Context> context, const LoggingOptions& loggingOptions) { |
| 1740 | // Set WebAssembly.Module @@HasInstance |
| 1741 | setWebAssemblyModuleHasInstance(lock, context); |
| 1742 | |
| 1743 | // Shim WebAssembly.instantiate to detect modules exporting "__instance_signal". |
| 1744 | if (util::Autogate::isEnabled(util::AutogateKey::WASM_SHUTDOWN_SIGNAL_SHIM)) { |
| 1745 | shimWebAssemblyInstantiate(lock, context); |
| 1746 | } |
| 1747 | |
| 1748 | // We replace the default V8 console.log(), etc. methods, to give the worker access to |
| 1749 | // logged content, and log formatted values to stdout/stderr locally. |
| 1750 | auto global = context->Global(); |
| 1751 | auto consoleStr = jsg::v8StrIntern(lock.v8Isolate, "console"); |
| 1752 | auto console = jsg::check(global->Get(context, consoleStr)).As<v8::Object>(); |
| 1753 | |
| 1754 | auto setHandler = [&](const char* method, LogLevel level) { |
| 1755 | auto methodStr = jsg::v8StrIntern(lock.v8Isolate, method); |
| 1756 | v8::Global<v8::Function> original( |
| 1757 | lock.v8Isolate, jsg::check(console->Get(context, methodStr)).As<v8::Function>()); |
| 1758 | |
| 1759 | auto f = lock.wrapSimpleFunction(context, |
| 1760 | [loggingOptions, level, original = kj::mv(original)]( |
| 1761 | jsg::Lock& js, const v8::FunctionCallbackInfo<v8::Value>& info) { |
| 1762 | handleLog(js, loggingOptions, level, original, info); |
| 1763 | }); |
| 1764 | jsg::check(console->Set(context, methodStr, f)); |
| 1765 | }; |
| 1766 | |
| 1767 | setHandler("debug", LogLevel::DEBUG_); |
| 1768 | setHandler("error", LogLevel::ERROR); |
| 1769 | setHandler("info", LogLevel::INFO); |
| 1770 | setHandler("log", LogLevel::LOG); |
| 1771 | setHandler("warn", LogLevel::WARN); |
| 1772 | } |
| 1773 | // ======================================================================================= |
| 1774 | |
| 1775 | namespace { |
| 1776 | kj::Maybe<jsg::JsObject> tryResolveMainModule(jsg::Lock& js, |
| 1777 | const kj::Path& mainModule, |
| 1778 | jsg::JsContext<api::ServiceWorkerGlobalScope>& jsContext, |
| 1779 | const Worker::Script& script, |
| 1780 | ExceptionOrDuration& limitErrorOrTime) { |
| 1781 | kj::Own<void> limitScope; |
| 1782 | if (script.isPython()) { |
| 1783 | limitScope = script.getIsolate().getLimitEnforcer().enterStartupPython(js, limitErrorOrTime); |
| 1784 | } else { |
| 1785 | limitScope = script.getIsolate().getLimitEnforcer().enterStartupJs(js, limitErrorOrTime); |
| 1786 | } |
| 1787 | |
| 1788 | KJ_DEFER({ |
| 1789 | if (limitErrorOrTime.is<kj::Exception>()) { |
| 1790 | // If we hit the limit in PerformMicrotaskCheckpoint() we may not have actually |
| 1791 | // thrown an exception. |
| 1792 | throw jsg::JsExceptionThrown(); |
| 1793 | } |
| 1794 | }); |
| 1795 | |
| 1796 | // Before resolving the main module, if both nodejs_compat_v2 and the new |
| 1797 | // module registry are enabled, let's pre-resolve the process and buffer modules. |
| 1798 | // Why? Great question! Resolving these modules synchronously causes the microtask |
| 1799 | // queue to be pumped, which we don't actually want to do while resolving the main |
| 1800 | // module until we are ready. Both process and buffer are exposed via globalThis |
| 1801 | // when the nodejs_compat_v2 flag is used, and if the top-level scope is accessing |
| 1802 | // either globalThis.process or globalThis.buffer, then we need to make sure that |
| 1803 | // the modules are already resolved so we don't pump the microtask queue while |
| 1804 | // synchronously accessing those globals. Resolving them here ensures that they are |
| 1805 | // ready to go before we begin evaluating the main module. |
| 1806 | auto featureFlags = FeatureFlags::get(js); |
| 1807 | if (featureFlags.getNodeJsCompatV2() && featureFlags.getNewModuleRegistry()) { |
| 1808 | JSG_REQUIRE_NONNULL(js.resolveModule("node:process", jsg::RequireEsm::YES), Error, |
| 1809 | "Failed to initialize node:process module"); |
| 1810 | JSG_REQUIRE_NONNULL(js.resolveModule("node:buffer", jsg::RequireEsm::YES), Error, |
| 1811 | "Failed to initialize node:buffer module"); |
| 1812 | } |
| 1813 | |
| 1814 | // When enable_nodejs_global_timers is enabled, load the module that makes all 6 timer |
| 1815 | // functions (setTimeout, setInterval, clearTimeout, clearInterval, setImmediate, |
| 1816 | // clearImmediate) available on globalThis as Node.js-compatible versions from node:timers. |
| 1817 | if (featureFlags.getEnableNodejsGlobalTimers()) { |
| 1818 | JSG_REQUIRE_NONNULL(js.resolveInternalModule("node-internal:internal_timers_global_override"), |
| 1819 | Error, "Failed to initialize node-internal:internal_timers_global_override module"); |
| 1820 | } |
| 1821 | |
| 1822 | return js.resolveModule(mainModule.toString(false), jsg::RequireEsm::YES); |
| 1823 | } |
| 1824 | } // anonymous namespace |
| 1825 | |
| 1826 | Worker::Worker(kj::Own<const Script> scriptParam, |
| 1827 | kj::Own<WorkerObserver> metricsParam, |
| 1828 | kj::FunctionParam<void(jsg::Lock& lock, |
| 1829 | const Api& api, |
| 1830 | v8::Local<v8::Object> target, |
| 1831 | v8::Local<v8::Object> ctxExports)> compileBindings, |
| 1832 | IsolateObserver::StartType startType, |
| 1833 | SpanParent parentSpan, |
| 1834 | LockType lockType, |
| 1835 | kj::Maybe<ValidationErrorReporter&> errorReporter, |
| 1836 | kj::Maybe<kj::Duration&> startupTime) |
| 1837 | : script(kj::mv(scriptParam)), |
| 1838 | metrics(kj::mv(metricsParam)), |
| 1839 | impl(kj::heap<Impl>()) { |
| 1840 | // Enter/lock isolate. |
| 1841 | jsg::runInV8Stack([&](jsg::V8StackScope& stackScope) { |
| 1842 | Isolate::Impl::Lock recordedLock(*script->isolate, lockType, stackScope); |
| 1843 | auto& lock = *recordedLock.lock; |
| 1844 | |
| 1845 | // If we throw an exception, it's important that `impl` is destroyed under lock. |
| 1846 | KJ_ON_SCOPE_FAILURE({ |
| 1847 | auto implToDestroy = kj::mv(impl); |
| 1848 | KJ_IF_SOME(c, implToDestroy->context) { |
| 1849 | recordedLock.disposeContext(kj::mv(c)); |
| 1850 | } else { |
| 1851 | // Else block to avoid dangling else clang warning. |
| 1852 | } |
| 1853 | }); |
| 1854 | |
| 1855 | auto maybeMakeSpan = [&](auto operationName) -> SpanBuilder { |
| 1856 | auto span = parentSpan.newChild(kj::mv(operationName)); |
| 1857 | if (span.isObserved()) { |
| 1858 | span.setTag("truncated_script_id"_kjc, truncateScriptId(script->getId())); |
| 1859 | } |
| 1860 | return span; |
| 1861 | }; |
| 1862 | |
| 1863 | auto currentSpan = maybeMakeSpan("lw:new_startup_metrics"_kjc); |
| 1864 | |
| 1865 | auto startupMetrics = metrics->startup(startType); |
| 1866 | |
| 1867 | currentSpan = maybeMakeSpan("lw:new_context"_kjc); |
| 1868 | |
| 1869 | // Create a stack-allocated handle scope. |
| 1870 | lock.withinHandleScope([&] { |
| 1871 | jsg::JsContext<api::ServiceWorkerGlobalScope>* jsContext; |
| 1872 | |
| 1873 | KJ_IF_SOME(c, script->impl->moduleContext) { |
| 1874 | // Use the shared context from the script. |
| 1875 | // const_cast OK because guarded by `lock`. |
| 1876 | jsContext = const_cast<jsg::JsContext<api::ServiceWorkerGlobalScope>*>(&c); |
| 1877 | currentSpan.setTag("module_context"_kjc, true); |
| 1878 | } else { |
| 1879 | // Create a new context. |
| 1880 | jsContext = &this->impl->context.emplace(script->isolate->getApi().newContext(lock, |
| 1881 | { |
| 1882 | .newModuleRegistry = script->impl->getNewModuleRegistry(), |
| 1883 | .schemaLoader = script->getSchemaLoader(), |
| 1884 | })); |
| 1885 | } |
| 1886 | |
| 1887 | v8::Local<v8::Context> context = KJ_REQUIRE_NONNULL(jsContext).getHandle(lock); |
| 1888 | |
| 1889 | // Install the virtual file system on the context. Keep in mind that for service |
| 1890 | // worker style workers, the Script may be shared between multiple Workers, even |
| 1891 | // across different accounts. Currently, the internal state of the VFS does not |
| 1892 | // contain any account-specific or worker-specific state so this is OK for now. |
| 1893 | // The VFS would contain the script files only and any temporary files created |
| 1894 | // within the context of a worker are always stored in temporary space attached |
| 1895 | // to the IoContext or the current execution context. If we extend these capabilities |
| 1896 | // in the future, we may need to revisit this. For modular workers, this is not |
| 1897 | // an issue since each Worker gets its own Script instance. |
| 1898 | script->installVirtualFileSystemOnContext(context); |
| 1899 | |
| 1900 | if (!script->modular) { |
| 1901 | recordedLock.setupContext(context); |
| 1902 | } |
| 1903 | |
| 1904 | if (script->impl->unboundScriptOrMainModule == nullptr) { |
| 1905 | // Script failed to parse. Act as if the script was empty -- i.e. do nothing. |
| 1906 | impl->permanentException = |
| 1907 | script->impl->permanentException.map([](auto& e) { return e.clone(); }); |
| 1908 | return; |
| 1909 | } |
| 1910 | |
| 1911 | // Enter the context for compiling and running the script. |
| 1912 | JSG_WITHIN_CONTEXT_SCOPE(lock, context, [&](jsg::Lock& js) { |
| 1913 | v8::TryCatch catcher(lock.v8Isolate); |
| 1914 | ExceptionOrDuration limitErrorOrTime = 0 * kj::NANOSECONDS; |
| 1915 | |
| 1916 | try { |
| 1917 | try { |
| 1918 | currentSpan = maybeMakeSpan("lw:globals_instantiation"_kjc); |
| 1919 | |
| 1920 | v8::Local<v8::Object> bindingsScope; |
| 1921 | if (script->isModular()) { |
| 1922 | // Use `env` variable. |
| 1923 | bindingsScope = v8::Object::New(lock.v8Isolate); |
| 1924 | if (!FeatureFlags::get(js).getDisableImportableEnv()) { |
| 1925 | lock.setWorkerEnv(lock.v8Ref(bindingsScope)); |
| 1926 | } |
| 1927 | } else { |
| 1928 | // Use global-scope bindings. |
| 1929 | bindingsScope = context->Global(); |
| 1930 | } |
| 1931 | |
| 1932 | // Load globals. |
| 1933 | // const_cast OK because we hold the lock. |
| 1934 | for (auto& global: const_cast<Script&>(*script).impl->globals) { |
| 1935 | lock.v8Set(bindingsScope, global.name, global.value); |
| 1936 | } |
| 1937 | |
| 1938 | v8::Local<v8::Object> ctxExports = v8::Object::New(lock.v8Isolate); |
| 1939 | |
| 1940 | compileBindings(lock, script->isolate->getApi(), bindingsScope, ctxExports); |
| 1941 | |
| 1942 | // Execute script. |
| 1943 | currentSpan = maybeMakeSpan("lw:top_level_execution"_kjc); |
| 1944 | |
| 1945 | // Ensure that our worker top-level bootstrap has a temporary directory |
| 1946 | // storage scope. This is used to store temporary files created within |
| 1947 | // the top-level evaluation of the worker. With this instantiated on |
| 1948 | // the stack, temporary files will be cleaned up when the scope is |
| 1949 | // destroyed, which means any temporary files created in the top-level |
| 1950 | // evaluation will *not* be available to the worker after the top-level |
| 1951 | // evaluation is complete. |
| 1952 | TmpDirStoreScope tmpDirStoreScope; |
| 1953 | |
| 1954 | // We allow eval and new Function() during startup, becaues startup time is entirely |
| 1955 | // deterministic, so we can easily reproduce the input to eval() by just running the |
| 1956 | // worker again. We do not allow eval() at runtime because we need to have a record of |
| 1957 | // all code that executes in production for forensic purposes, and at runtime the input |
| 1958 | // to eval() could have come from a remote source on which we don't have a record. |
| 1959 | js.setAllowEval(FeatureFlags::get(js).getAllowEvalDuringStartup()); |
| 1960 | KJ_DEFER(js.setAllowEval(false)); |
| 1961 | |
| 1962 | KJ_SWITCH_ONEOF(script->impl->unboundScriptOrMainModule) { |
| 1963 | KJ_CASE_ONEOF(unboundScript, jsg::NonModuleScript) { |
| 1964 | auto limitScope = |
| 1965 | script->isolate->getLimitEnforcer().enterStartupJs(lock, limitErrorOrTime); |
| 1966 | unboundScript.run(lock); |
| 1967 | // Flush microtasks enqueued during top-level script evaluation. |
| 1968 | // Without this flush, microtasks (e.g. promise continuations from async |
| 1969 | // initialization) remain on the per-isolate microtask queue and can leak across |
| 1970 | // V8 contexts when multiple Workers share an isolate (same script, different |
| 1971 | // zones). The leaked microtasks then execute under the wrong IoContext, making |
| 1972 | // things go boom. |
| 1973 | lock.runMicrotasks(); |
| 1974 | } |
| 1975 | KJ_CASE_ONEOF(mainModule, kj::Path) { |
| 1976 | KJ_IF_SOME(ns, |
| 1977 | tryResolveMainModule(lock, mainModule, *jsContext, *script, limitErrorOrTime)) { |
| 1978 | impl->env = lock.v8Ref(bindingsScope.As<v8::Value>()); |
| 1979 | impl->ctxExports = lock.v8Ref(ctxExports.As<v8::Value>()); |
| 1980 | |
| 1981 | if (!FeatureFlags::get(js).getDisableImportableEnv()) { |
| 1982 | lock.setWorkerExports(lock.v8Ref(ctxExports)); |
| 1983 | } |
| 1984 | |
| 1985 | auto& api = script->isolate->getApi(); |
| 1986 | auto handlers = api.unwrapExports(lock, ns); |
| 1987 | auto entrypointClasses = api.getEntrypointClasses(lock); |
| 1988 | |
| 1989 | for (auto& handler: handlers.fields) { |
| 1990 | KJ_SWITCH_ONEOF(handler.value) { |
| 1991 | KJ_CASE_ONEOF(obj, api::ExportedHandler) { |
| 1992 | obj.env = lock.v8Ref(bindingsScope.As<v8::Value>()); |
| 1993 | // Historically, non-class-based handlers reused the same ctx object for all requests. |
| 1994 | // This was an accident, but some Workers depend on it. |
| 1995 | // Newer worker with the unique_ctx_per_invocation will allocate a new ctx for every request. |
| 1996 | obj.ctx = js.alloc<api::ExecutionContext>(lock, jsg::JsValue(ctxExports)); |
| 1997 | |
| 1998 | // Python Workers append all durable objects, worker entrypoint and workflow |
| 1999 | // entrypoint classes in the pythonEntrypoints named export. |
| 2000 | bool isPythonWorker = FeatureFlags::get(js).getPythonWorkers(); |
| 2001 | if (handler.name == "pythonEntrypoints" && isPythonWorker) { |
| 2002 | auto handle = obj.self.getHandle(js); |
| 2003 | auto dict = js.toDict(handle); |
| 2004 | for (auto& field: dict.fields) { |
| 2005 | auto unwrapped = api.unwrapExport(lock, field.value); |
| 2006 | KJ_SWITCH_ONEOF(unwrapped) { |
| 2007 | KJ_CASE_ONEOF(cls, EntrypointClass) { |
| 2008 | processEntrypointClass( |
| 2009 | js, kj::mv(cls), entrypointClasses, kj::mv(field.name)); |
| 2010 | } |
| 2011 | KJ_CASE_ONEOF(obj, api::ExportedHandler) { |
| 2012 | KJ_FAIL_ASSERT("Expected EntrypointClass"); |
| 2013 | } |
| 2014 | } |
| 2015 | } |
| 2016 | } else { |
| 2017 | impl->namedHandlers.insert(kj::mv(handler.name), kj::mv(obj)); |
| 2018 | } |
| 2019 | } |
| 2020 | KJ_CASE_ONEOF(cls, EntrypointClass) { |
| 2021 | processEntrypointClass( |
| 2022 | js, kj::mv(cls), entrypointClasses, kj::mv(handler.name)); |
| 2023 | } |
| 2024 | } |
| 2025 | } |
| 2026 | } else { |
| 2027 | JSG_FAIL_REQUIRE(TypeError, "Main module name is not present in bundle."); |
| 2028 | } |
| 2029 | } |
| 2030 | } |
| 2031 | |
| 2032 | KJ_IF_SOME(s, startupTime) { |
| 2033 | KJ_SWITCH_ONEOF(limitErrorOrTime) { |
| 2034 | KJ_CASE_ONEOF(startupTimeElapsed, kj::Duration) { |
| 2035 | s = startupTimeElapsed; |
| 2036 | } |
| 2037 | KJ_CASE_ONEOF(limitError, kj::Exception) {} |
| 2038 | } |
| 2039 | } else { |
| 2040 | } |
| 2041 | startupMetrics->done(); |
| 2042 | } catch (const kj::Exception& e) { |
| 2043 | lock.throwException(e.clone()); |
| 2044 | // lock.throwException() here will throw a jsg::JsExceptionThrown which we catch |
| 2045 | // in the outer try/catch. |
| 2046 | } |
| 2047 | } catch (const jsg::JsExceptionThrown&) { |
| 2048 | reportStartupError(script->id, lock, script->isolate->impl->inspector, |
| 2049 | script->isolate->getLimitEnforcer(), kj::mv(limitErrorOrTime), catcher, errorReporter, |
| 2050 | impl->permanentException, currentSpan, script->getDynamicEnvBuilder() != kj::none); |
| 2051 | } |
| 2052 | }); |
| 2053 | |
| 2054 | // Reset this back to its default after startup execution |
| 2055 | // Leaving it on comes at the expense of collecting stack traces for all thrown exceptions |
| 2056 | // Ref: https://github.com/cloudflare/workerd/issues/5332 |
| 2057 | if (script->isolate->impl->inspector == kj::none) { |
| 2058 | lock.v8Isolate->SetCaptureStackTraceForUncaughtExceptions(false); |
| 2059 | } |
| 2060 | }); |
| 2061 | }); |
| 2062 | } |
| 2063 | |
| 2064 | Worker::~Worker() noexcept(false) { |
| 2065 | metrics->teardownStarted(); |
| 2066 | |
| 2067 | auto& isolateImpl = *script->getIsolate().impl; |
| 2068 | auto lock = isolateImpl.workerDestructionQueue.lockExclusive(); |
| 2069 | |
| 2070 | // Previously, this metric meant the isolate lock. We might as well make it mean the worker |
| 2071 | // destruction queue lock now to verify it is much less-contended than the isolate lock. |
| 2072 | metrics->teardownLockAcquired(); |
| 2073 | |
| 2074 | // Defer destruction of our V8 objects, in particular our jsg::Context, which requires some |
| 2075 | // finalization. |
| 2076 | lock->push(kj::mv(impl)); |
| 2077 | } |
| 2078 | |
| 2079 | void Worker::processEntrypointClass(jsg::Lock& js, |
| 2080 | EntrypointClass cls, |
| 2081 | EntrypointClasses entrypointClasses, |
| 2082 | kj::String handlerName) { |
| 2083 | js.withinHandleScope([&]() { |
| 2084 | jsg::JsObject handle(KJ_ASSERT_NONNULL(cls.tryGetHandle(js.v8Isolate))); |
| 2085 | |
| 2086 | for (;;) { |
| 2087 | if (handle == entrypointClasses.durableObject) { |
| 2088 | impl->actorClasses.insert(kj::mv(handlerName), |
| 2089 | ActorClassInfo{ |
| 2090 | .cls = kj::mv(cls), |
| 2091 | .missingSuperclass = false, |
| 2092 | }); |
| 2093 | return; |
| 2094 | } else if (handle == entrypointClasses.workerEntrypoint) { |
| 2095 | impl->statelessClasses.insert(kj::mv(handlerName), kj::mv(cls)); |
| 2096 | return; |
| 2097 | } else if (handle == entrypointClasses.workflowEntrypoint) { |
| 2098 | impl->workflowClasses.insert(kj::mv(handlerName), kj::mv(cls)); |
| 2099 | return; |
| 2100 | } |
| 2101 | |
| 2102 | handle = KJ_UNWRAP_OR(handle.getPrototype(js).tryCast<jsg::JsObject>(), { |
| 2103 | // Reached end of prototype chain. |
| 2104 | |
| 2105 | // For historical reasons, we assume a class is a Durable Object |
| 2106 | // class if it doesn't inherit anything. |
| 2107 | // TODO(someday): Log a warning suggesting extending DurableObject. |
| 2108 | // TODO(someday): Introduce a compat flag that makes this required. |
| 2109 | impl->actorClasses.insert(kj::mv(handlerName), |
| 2110 | ActorClassInfo{ |
| 2111 | .cls = kj::mv(cls), |
| 2112 | .missingSuperclass = true, |
| 2113 | }); |
| 2114 | return; |
| 2115 | }); |
| 2116 | } |
| 2117 | }); |
| 2118 | } |
| 2119 | |
| 2120 | void Worker::handleLog(jsg::Lock& js, |
| 2121 | const LoggingOptions& loggingOptions, |
| 2122 | LogLevel level, |
| 2123 | const v8::Global<v8::Function>& original, |
| 2124 | const v8::FunctionCallbackInfo<v8::Value>& info) { |
| 2125 | // Call original V8 implementation so messages sent to connected inspector if any |
| 2126 | auto context = js.v8Context(); |
| 2127 | int length = info.Length(); |
| 2128 | // to pass additional arguments from this function to js' `formatLog` we add arguments to the end |
| 2129 | // of the arguments vector, then in formatLog we `pop` these from the vector. |
| 2130 | // 3 is just the number of args we currently pass. |
| 2131 | v8::LocalVector<v8::Value> args(js.v8Isolate, length + 3); |
| 2132 | for (auto i: kj::zeroTo(length)) args[i] = info[i]; |
| 2133 | jsg::check(original.Get(js.v8Isolate)->Call(context, info.This(), length, args.data())); |
| 2134 | |
| 2135 | // The TryCatch is initialized here to catch cases where the v8 isolate's execution is |
| 2136 | // terminating, usually as a result of an infinite loop. We need to perform the initialization |
| 2137 | // here because `message` is called multiple times. |
| 2138 | v8::TryCatch tryCatch(js.v8Isolate); |
| 2139 | auto message = [&]() { |
| 2140 | int length = info.Length(); |
| 2141 | kj::Vector<kj::String> stringified(length); |
| 2142 | for (auto i: kj::zeroTo(length)) { |
| 2143 | auto arg = info[i]; |
| 2144 | // serializeJson and v8::Value::ToString can throw JS exceptions |
| 2145 | // (e.g. for recursive objects) so we eat them here, to ensure logging and non-logging code |
| 2146 | // have the same exception behavior. |
| 2147 | if (!tryCatch.CanContinue()) { |
| 2148 | stringified.add(kj::str("{}")); |
| 2149 | break; |
| 2150 | } |
| 2151 | // The following code checks the `arg` to see if it should be serialised to JSON. |
| 2152 | // |
| 2153 | // We use the following criteria: if arg is null, a number, a boolean, an array, a string, an |
| 2154 | // object or it defines a `toJSON` property that is a function, then the arg gets serialised |
| 2155 | // to JSON. |
| 2156 | // |
| 2157 | // Otherwise we stringify the argument. |
| 2158 | js.withinHandleScope([&] { |
| 2159 | auto context = js.v8Context(); |
| 2160 | bool shouldSerialiseToJson = false; |
| 2161 | if (arg->IsNull() || arg->IsNumber() || arg->IsArray() || arg->IsBoolean() || |
| 2162 | arg->IsString() || |
| 2163 | arg->IsUndefined()) { // This is special cased for backwards compatibility. |
| 2164 | shouldSerialiseToJson = true; |
| 2165 | } |
| 2166 | if (arg->IsObject()) { |
| 2167 | v8::Local<v8::Object> obj = arg.As<v8::Object>(); |
| 2168 | v8::Local<v8::Object> freshObj = v8::Object::New(js.v8Isolate); |
| 2169 | |
| 2170 | // Determine whether `obj` is constructed using `{}` or `new Object()`. This ensures |
| 2171 | // we don't serialise values like Promises to JSON. |
| 2172 | #if V8_MAJOR_VERSION >= 15 || (V8_MAJOR_VERSION == 14 && V8_MINOR_VERSION >= 7) |
| 2173 | if (obj->GetPrototype()->SameValue(freshObj->GetPrototype()) || |
| 2174 | obj->GetPrototype()->IsNull()) { |
| 2175 | #else |
| 2176 | // TODO(cleanup): Remove when unnecessary. |
| 2177 | if (obj->GetPrototypeV2()->SameValue(freshObj->GetPrototypeV2()) || |
| 2178 | obj->GetPrototypeV2()->IsNull()) { |
| 2179 | #endif |
| 2180 | shouldSerialiseToJson = true; |
| 2181 | } |
| 2182 | |
| 2183 | // Check if arg has a `toJSON` property which is a function. |
| 2184 | auto toJSONStr = jsg::v8StrIntern(js.v8Isolate, "toJSON"_kj); |
| 2185 | v8::MaybeLocal<v8::Value> toJSON = obj->GetRealNamedProperty(context, toJSONStr); |
| 2186 | if (!toJSON.IsEmpty()) { |
| 2187 | if (jsg::check(toJSON)->IsFunction()) { |
| 2188 | shouldSerialiseToJson = true; |
| 2189 | } |
| 2190 | } |
| 2191 | } |
| 2192 | |
| 2193 | if (kj::runCatchingExceptions([&]() { |
| 2194 | // On the off chance the the arg is the request.cf object, let's make |
| 2195 | // sure we do not log proxied fields here. |
| 2196 | if (shouldSerialiseToJson) { |
| 2197 | auto s = js.serializeJson(arg); |
| 2198 | // serializeJson returns the string "undefined" for some values (undefined, |
| 2199 | // Symbols, functions). We remap these values to null to ensure valid JSON output. |
| 2200 | if (s == "undefined"_kj) { |
| 2201 | stringified.add(kj::str("null")); |
| 2202 | } else { |
| 2203 | stringified.add(kj::mv(s)); |
| 2204 | } |
| 2205 | } else { |
| 2206 | stringified.add(js.serializeJson(jsg::check(arg->ToString(context)))); |
| 2207 | } |
| 2208 | }) != kj::none) { |
| 2209 | stringified.add(kj::str("{}")); |
| 2210 | }; |
| 2211 | }); |
| 2212 | } |
| 2213 | return kj::str("[", kj::delimited(stringified, ", "_kj), "]"); |
| 2214 | }; |
| 2215 | |
| 2216 | // Only check tracing if console.log() was not invoked at the top level. |
| 2217 | KJ_IF_SOME(ioContext, IoContext::tryCurrent()) { |
| 2218 | KJ_IF_SOME(tracer, ioContext.getWorkerTracer()) { |
| 2219 | auto timestamp = ioContext.now(); |
| 2220 | tracer.addLog(ioContext.getInvocationSpanContext(), timestamp, level, message()); |
| 2221 | } |
| 2222 | } |
| 2223 | |
| 2224 | if (loggingOptions.consoleMode == Worker::ConsoleMode::INSPECTOR_ONLY) { |
| 2225 | // Lets us dump console.log()s to stdout when running test-runner with --verbose flag, to make |
| 2226 | // it easier to debug tests. Note that when --verbose is not passed, KJ_LOG(INFO, ...) will |
| 2227 | // not even evaluate its arguments, so `message()` will not be called at all. |
| 2228 | KJ_LOG(INFO, "console.log()", message()); |
| 2229 | } else { |
| 2230 | // Write to stdio if allowed by console mode. This is making use of our internal |
| 2231 | // built-in implementation of the node:util inspect API. |
| 2232 | static const ColorMode COLOR_MODE = permitsColor(); |
| 2233 | #if _WIN32 |
| 2234 | static bool STDOUT_TTY = _isatty(_fileno(stdout)); |
| 2235 | static bool STDERR_TTY = _isatty(_fileno(stderr)); |
| 2236 | #else |
| 2237 | static bool STDOUT_TTY = isatty(STDOUT_FILENO); |
| 2238 | static bool STDERR_TTY = isatty(STDERR_FILENO); |
| 2239 | #endif |
| 2240 | |
| 2241 | // Log warnings and errors to stderr |
| 2242 | // Always log to stdout when structuredLogging is enabled. |
| 2243 | auto useStderr = level >= LogLevel::WARN && !loggingOptions.structuredLogging; |
| 2244 | auto fd = useStderr ? stderr : stdout; |
| 2245 | auto tty = useStderr ? STDERR_TTY : STDOUT_TTY; |
| 2246 | auto colors = |
| 2247 | COLOR_MODE == ColorMode::ENABLED || (COLOR_MODE == ColorMode::ENABLED_IF_TTY && tty); |
| 2248 | |
| 2249 | constexpr auto kSpecifier = "node-internal:internal_inspect"_kj; |
| 2250 | auto inspectModule = KJ_ASSERT_NONNULL(js.resolveInternalModule(kSpecifier)); |
| 2251 | v8::Local<v8::Value> formatLogVal = inspectModule.get(js, "formatLog"_kj); |
| 2252 | KJ_ASSERT(formatLogVal->IsFunction()); |
| 2253 | auto formatLog = formatLogVal.As<v8::Function>(); |
| 2254 | |
| 2255 | auto levelStr = logLevelToString(level); |
| 2256 | args[length] = js.boolean(colors); |
| 2257 | args[length + 1] = js.boolean(loggingOptions.structuredLogging.toBool()); |
| 2258 | args[length + 2] = js.strIntern(levelStr); |
| 2259 | auto formatted = js.toString( |
| 2260 | jsg::check(formatLog->Call(context, js.v8Undefined(), length + 3, args.data()))); |
| 2261 | fprintf(fd, "%s\n", formatted.cStr()); |
| 2262 | fflush(fd); |
| 2263 | } |
| 2264 | } |
| 2265 | |
| 2266 | Worker::Lock::TakeSynchronously::TakeSynchronously(kj::Maybe<RequestObserver&> requestParam) { |
| 2267 | KJ_IF_SOME(r, requestParam) { |
| 2268 | request = &r; |
| 2269 | } |
| 2270 | } |
| 2271 | |
| 2272 | kj::Maybe<RequestObserver&> Worker::Lock::TakeSynchronously::getRequest() { |
| 2273 | if (request != nullptr) { |
| 2274 | return *request; |
| 2275 | } |
| 2276 | return kj::none; |
| 2277 | } |
| 2278 | |
| 2279 | struct Worker::Lock::Impl { |
| 2280 | Isolate::Impl::Lock recordedLock; |
| 2281 | jsg::Lock& inner; |
| 2282 | |
| 2283 | Impl(const Worker& worker, LockType lockType, jsg::V8StackScope& stackScope) |
| 2284 | : recordedLock(worker.getIsolate(), lockType, stackScope), |
| 2285 | inner(*recordedLock.lock) {} |
| 2286 | }; |
| 2287 | |
| 2288 | Worker::Lock::Lock(const Worker& constWorker, LockType lockType, jsg::V8StackScope& stackScope) |
| 2289 | : // const_cast OK because we took out a lock. |
| 2290 | worker(const_cast<Worker&>(constWorker)), |
| 2291 | impl(kj::heap<Impl>(worker, lockType, stackScope)) { |
| 2292 | kj::requireOnStack(this, "Worker::Lock MUST be allocated on the stack."); |
| 2293 | } |
| 2294 | |
| 2295 | Worker::Lock::~Lock() noexcept(false) { |
| 2296 | // const_cast OK because we hold -- nay, we *are* -- a lock on the script. |
| 2297 | auto& isolate = const_cast<Isolate&>(worker.getIsolate()); |
| 2298 | if (impl->recordedLock.checkInWithLimitEnforcer(isolate)) { |
| 2299 | isolate.disconnectInspector(); |
| 2300 | } |
| 2301 | } |
| 2302 | |
| 2303 | void Worker::Lock::requireNoPermanentException() { |
| 2304 | KJ_IF_SOME(e, worker.impl->permanentException) { |
| 2305 | // Block taking lock when worker failed to start up. |
| 2306 | kj::throwFatalException(e.clone()); |
| 2307 | } |
| 2308 | } |
| 2309 | |
| 2310 | Worker::Lock::operator jsg::Lock&() { |
| 2311 | return impl->inner; |
| 2312 | } |
| 2313 | |
| 2314 | v8::Isolate* Worker::Lock::getIsolate() { |
| 2315 | return impl->inner.v8Isolate; |
| 2316 | } |
| 2317 | |
| 2318 | v8::Local<v8::Context> Worker::Lock::getContext() { |
| 2319 | KJ_IF_SOME(c, worker.impl->context) { |
| 2320 | return c.getHandle(impl->inner); |
| 2321 | } else KJ_IF_SOME(c, const_cast<Script&>(*worker.script).impl->moduleContext) { |
| 2322 | return c.getHandle(impl->inner); |
| 2323 | } else { |
| 2324 | KJ_UNREACHABLE; |
| 2325 | } |
| 2326 | } |
| 2327 | |
| 2328 | template <typename T> |
| 2329 | static inline kj::Own<T> fakeOwn(T& ref) { |
| 2330 | return kj::Own<T>(&ref, kj::NullDisposer::instance); |
| 2331 | } |
| 2332 | |
| 2333 | kj::Maybe<kj::Own<api::ExportedHandler>> Worker::Lock::getExportedHandler( |
| 2334 | kj::Maybe<kj::StringPtr> name, |
| 2335 | kj::Maybe<VersionInfo> versionInfo, |
| 2336 | Frankenvalue props, |
| 2337 | kj::Maybe<Worker::Actor&> actor, |
| 2338 | bool isDynamicDispatch) { |
| 2339 | KJ_IF_SOME(a, actor) { |
| 2340 | KJ_IF_SOME(h, a.getHandler()) { |
| 2341 | return fakeOwn(h); |
| 2342 | } |
| 2343 | } |
| 2344 | |
| 2345 | kj::StringPtr n = name.orDefault("default"_kj); |
| 2346 | |
| 2347 | auto getHandlerFromEntrypointClass = |
| 2348 | [&](EntrypointClass& cls) -> kj::Maybe<kj::Own<api::ExportedHandler>> { |
| 2349 | jsg::Lock& js = *this; |
| 2350 | auto handler = kj::heap(cls(js, |
| 2351 | js.alloc<api::ExecutionContext>(js, |
| 2352 | jsg::JsValue(KJ_ASSERT_NONNULL(worker.impl->ctxExports).getHandle(js)), props.toJs(js), |
| 2353 | kj::mv(versionInfo)), |
| 2354 | KJ_ASSERT_NONNULL(worker.impl->env).addRef(js))); |
| 2355 | |
| 2356 | // HACK: We set handler.env and handler.ctx to undefined because we already passed the real |
| 2357 | // env and ctx into the constructor, and we want the handler methods to act like they take |
| 2358 | // just one parameter. |
| 2359 | handler->env = js.v8Ref(js.v8Undefined()); |
| 2360 | handler->ctx = kj::none; |
| 2361 | |
| 2362 | return handler; |
| 2363 | }; |
| 2364 | |
| 2365 | KJ_IF_SOME(h, worker.impl->namedHandlers.find(n)) { |
| 2366 | jsg::Lock& js = *this; |
| 2367 | if (!FeatureFlags::get(js).getReuseCtxAcrossNonclassEvents()) { |
| 2368 | api::ExportedHandler constructedHandler = h.clone(js); |
| 2369 | constructedHandler.ctx = js.alloc<api::ExecutionContext>(js, |
| 2370 | jsg::JsValue(KJ_ASSERT_NONNULL(worker.impl->ctxExports).getHandle(js)), props.toJs(js), |
| 2371 | kj::mv(versionInfo)); |
| 2372 | return kj::heap(kj::mv(constructedHandler)); |
| 2373 | } |
| 2374 | return fakeOwn(h); |
| 2375 | } else KJ_IF_SOME(cls, worker.impl->statelessClasses.find(n)) { |
| 2376 | return getHandlerFromEntrypointClass(cls); |
| 2377 | } else KJ_IF_SOME(cls, worker.impl->workflowClasses.find(n)) { |
| 2378 | return getHandlerFromEntrypointClass(cls); |
| 2379 | } else if (name == kj::none) { |
| 2380 | // If the default export was requested, and we didn't find a handler for it, we'll fall back |
| 2381 | // to addEventListener(). |
| 2382 | // |
| 2383 | // Note: The original intention was that we only use addEventListener() for |
| 2384 | // service-worker-syntax scripts, but apparently the code has long allowed it for |
| 2385 | // modules-based script too, if they lacked an `export default`. Yikes! Sadly, there are |
| 2386 | // Workers in production relying on this so we are stuck with it. |
| 2387 | return kj::none; |
| 2388 | } else { |
| 2389 | if (worker.impl->actorClasses.find(n) != kj::none) { |
| 2390 | if (isDynamicDispatch) { |
| 2391 | JSG_FAIL_REQUIRE(TypeError, "The entrypoint name ", n, |
| 2392 | " refers to a Durable Object class, but the incoming request is trying to invoke it as" |
| 2393 | " a stateless worker."); |
| 2394 | } else { |
| 2395 | LOG_ERROR_PERIODICALLY("worker is not an actor but class name was requested", n); |
| 2396 | } |
| 2397 | } else if (isDynamicDispatch) { |
| 2398 | JSG_FAIL_REQUIRE(TypeError, "The entrypoint name ", n, |
| 2399 | " was not found in this worker. Ensure the worker exports an entrypoint with that name."); |
| 2400 | } else { |
| 2401 | LOG_ERROR_PERIODICALLY("worker has no such named entrypoint", n); |
| 2402 | } |
| 2403 | |
| 2404 | KJ_FAIL_ASSERT("worker_do_not_log; Unable to get exported handler"); |
| 2405 | }; |
| 2406 | } |
| 2407 | |
| 2408 | api::ServiceWorkerGlobalScope& Worker::Lock::getGlobalScope() { |
| 2409 | return KJ_ASSERT_NONNULL(jsg::getAlignedPointerFromEmbedderData<api::ServiceWorkerGlobalScope>( |
| 2410 | getContext(), jsg::ContextPointerSlot::GLOBAL_WRAPPER)); |
| 2411 | } |
| 2412 | |
| 2413 | TimeoutId::Generator& Worker::Lock::getTimeoutIdGenerator() { |
| 2414 | return getGlobalScope().timeoutIdGenerator; |
| 2415 | } |
| 2416 | |
| 2417 | jsg::AsyncContextFrame::StorageKey& Worker::Lock::getTraceAsyncContextKey() { |
| 2418 | // const_cast OK because we are a lock on this isolate. |
| 2419 | auto& isolate = const_cast<Isolate&>(worker.getIsolate()); |
| 2420 | return *(isolate.traceAsyncContextKey); |
| 2421 | } |
| 2422 | |
| 2423 | jsg::AsyncContextFrame::StorageKey& Worker::Lock::getUserTraceAsyncContextKey() { |
| 2424 | // const_cast OK because we are a lock on this isolate. |
| 2425 | auto& isolate = const_cast<Isolate&>(worker.getIsolate()); |
| 2426 | return *(isolate.userTraceAsyncContextKey); |
| 2427 | } |
| 2428 | |
| 2429 | bool Worker::Lock::isInspectorEnabled() { |
| 2430 | return worker.script->isolate->impl->inspector != kj::none; |
| 2431 | } |
| 2432 | |
| 2433 | void Worker::Lock::logWarning(kj::StringPtr description) { |
| 2434 | // const_cast OK because we are a lock on this isolate. |
| 2435 | const_cast<Isolate&>(worker.getIsolate()).logWarning(description, *this); |
| 2436 | } |
| 2437 | |
| 2438 | void Worker::Lock::logWarningOnce(kj::StringPtr description) { |
| 2439 | // const_cast OK because we are a lock on this isolate. |
| 2440 | const_cast<Isolate&>(worker.getIsolate()).logWarningOnce(description, *this); |
| 2441 | } |
| 2442 | |
| 2443 | void Worker::Lock::logErrorOnce(kj::StringPtr description) { |
| 2444 | // const_cast OK because we are a lock on this isolate. |
| 2445 | const_cast<Isolate&>(worker.getIsolate()).logErrorOnce(description); |
| 2446 | } |
| 2447 | |
| 2448 | void Worker::Lock::logUncaughtException(kj::StringPtr description) { |
| 2449 | // We don't add the exception to traces here, since it turns out that this path only gets hit by |
| 2450 | // intermediate exception handling. |
| 2451 | KJ_IF_SOME(i, worker.script->isolate->impl->inspector) { |
| 2452 | JSG_WITHIN_CONTEXT_SCOPE(*this, getContext(), |
| 2453 | [&](jsg::Lock& js) { jsg::sendExceptionToInspector(js, *i.get(), description); }); |
| 2454 | } |
| 2455 | |
| 2456 | // Run with --verbose to log JS exceptions to stderr. Useful when running tests. |
| 2457 | KJ_LOG(INFO, "uncaught exception", description); |
| 2458 | } |
| 2459 | |
| 2460 | void Worker::Lock::logUncaughtException( |
| 2461 | UncaughtExceptionSource source, const jsg::JsValue& exception, const jsg::JsMessage& message) { |
| 2462 | // Only add exception to trace when running within an I/O context with a tracer. |
| 2463 | KJ_IF_SOME(ioContext, IoContext::tryCurrent()) { |
| 2464 | KJ_IF_SOME(tracer, ioContext.getWorkerTracer()) { |
| 2465 | JSG_WITHIN_CONTEXT_SCOPE(*this, getContext(), [&](jsg::Lock& js) { |
| 2466 | addExceptionToTrace(impl->inner, ioContext, tracer, source, exception, |
| 2467 | worker.getIsolate().getApi().getErrorInterfaceTypeHandler(*this)); |
| 2468 | }); |
| 2469 | } |
| 2470 | } |
| 2471 | |
| 2472 | KJ_IF_SOME(i, worker.script->isolate->impl->inspector) { |
| 2473 | JSG_WITHIN_CONTEXT_SCOPE(*this, getContext(), |
| 2474 | [&](jsg::Lock& js) { sendExceptionToInspector(js, *i.get(), source, exception, message); }); |
| 2475 | } |
| 2476 | |
| 2477 | // Run with --verbose to log JS exceptions to stderr. Useful when running tests. |
| 2478 | if (kj::_::Debug::shouldLog(::kj::LogSeverity::INFO)) { |
| 2479 | JSG_WITHIN_CONTEXT_SCOPE(*this, getContext(), [&](jsg::Lock& js) { |
| 2480 | // Try to log `error.stack` if it exists. |
| 2481 | KJ_IF_SOME(obj, exception.tryCast<jsg::JsObject>()) { |
| 2482 | auto stack = obj.get(js, "stack"); |
| 2483 | if (!stack.isUndefined()) { |
| 2484 | KJ_LOG(INFO, "uncaught exception", source, stack); |
| 2485 | return; |
| 2486 | } |
| 2487 | } else { |
| 2488 | // Compiler gives a spurious warning if this `else` isn't here. |
| 2489 | } |
| 2490 | |
| 2491 | KJ_LOG(INFO, "uncaught exception", source, exception); |
| 2492 | }); |
| 2493 | } |
| 2494 | } |
| 2495 | |
| 2496 | void Worker::Lock::logUncaughtException(UncaughtExceptionSource source, kj::Exception&& exception) { |
| 2497 | jsg::Lock& js = *this; |
| 2498 | try { |
| 2499 | auto jsError = js.exceptionToJsValue(kj::mv(exception), |
| 2500 | { |
| 2501 | .trusted = true, |
| 2502 | }); |
| 2503 | logUncaughtException(source, jsError.getHandle(js)); |
| 2504 | } catch (const jsg::JsExceptionThrown&) { |
| 2505 | // An exception occurred while trying to convert the exception to a JS value. |
| 2506 | // With exceptionToJs, this should only happen if the isolate is terminating |
| 2507 | // because of a fatal error when trying to deserialize a tunneled exception |
| 2508 | // detail. In this case, we will want to log the original exception instead, |
| 2509 | // so let's try exceptionToJs again but this time ignoring the detail, and |
| 2510 | // if it throws again, we'll give up and propagate that exception to the |
| 2511 | // caller. |
| 2512 | auto jsError = js.exceptionToJsValue(exception.clone(), {.ignoreDetail = true}); |
| 2513 | logUncaughtException(source, jsError.getHandle(js)); |
| 2514 | } |
| 2515 | } |
| 2516 | |
| 2517 | void Worker::Lock::reportPromiseRejectEvent(v8::PromiseRejectMessage& message) { |
| 2518 | getGlobalScope().emitPromiseRejection(*this, message.GetEvent(), |
| 2519 | jsg::V8Ref<v8::Promise>(getIsolate(), message.GetPromise()), |
| 2520 | jsg::V8Ref<v8::Value>(getIsolate(), message.GetValue())); |
| 2521 | } |
| 2522 | |
| 2523 | void Worker::Lock::validateHandlers(ValidationErrorReporter& errorReporter) { |
| 2524 | JSG_WITHIN_CONTEXT_SCOPE(*this, getContext(), [&](jsg::Lock& js) { |
| 2525 | kj::HashSet<kj::StringPtr> ignoredHandlers; |
| 2526 | ignoredHandlers.insert("alarm"_kj); |
| 2527 | ignoredHandlers.insert("unhandledrejection"_kj); |
| 2528 | ignoredHandlers.insert("rejectionhandled"_kj); |
| 2529 | |
| 2530 | // Helper function to collect methods from a prototype chain |
| 2531 | auto collectMethodsFromPrototypeChain = [&](jsg::JsValue startProto, |
| 2532 | kj::HashSet<kj::String>& seenNames) { |
| 2533 | // Find the prototype for `Object` by creating one. |
| 2534 | auto obj = js.obj(); |
| 2535 | jsg::JsValue prototypeOfObject = obj.getPrototype(js); |
| 2536 | |
| 2537 | // Walk the prototype chain. |
| 2538 | jsg::JsValue proto = startProto; |
| 2539 | for (;;) { |
| 2540 | auto protoObj = KJ_UNWRAP_OR(proto.tryCast<jsg::JsObject>(), { |
| 2541 | errorReporter.addError( |
| 2542 | kj::str("Exported value's prototype chain does not end in Object.")); |
| 2543 | return; |
| 2544 | }); |
| 2545 | if (protoObj == prototypeOfObject) { |
| 2546 | // Reached the prototype for `Object`. Stop here. |
| 2547 | break; |
| 2548 | } |
| 2549 | |
| 2550 | // Awkwardly, the prototype's members are not typically enumerable, so we have to |
| 2551 | // enumerate them rather directly. |
| 2552 | jsg::JsArray properties = protoObj.getPropertyNames(js, jsg::KeyCollectionFilter::OWN_ONLY, |
| 2553 | jsg::PropertyFilter::SKIP_SYMBOLS, jsg::IndexFilter::SKIP_INDICES); |
| 2554 | for (auto i: kj::zeroTo(properties.size())) { |
| 2555 | auto name = properties.get(js, i).toString(js); |
| 2556 | if (name == "constructor"_kj) { |
| 2557 | // Don't treat special method `constructor` as an exported handler. |
| 2558 | continue; |
| 2559 | } |
| 2560 | |
| 2561 | if (!ignoredHandlers.contains(name)) { |
| 2562 | // Only report each method name once, even if it overrides a method in a superclass. |
| 2563 | seenNames.upsert(kj::mv(name), [&](auto&, auto&&) {}); |
| 2564 | } |
| 2565 | } |
| 2566 | |
| 2567 | proto = protoObj.getPrototype(js); |
| 2568 | } |
| 2569 | }; |
| 2570 | |
| 2571 | KJ_IF_SOME(c, worker.impl->context) { |
| 2572 | // Service workers syntax. |
| 2573 | auto handlerNames = c->getHandlerNames(); |
| 2574 | kj::Vector<kj::String> handlers; |
| 2575 | for (auto& name: handlerNames) { |
| 2576 | if (!ignoredHandlers.contains(name)) { |
| 2577 | handlers.add(kj::str(name)); |
| 2578 | } |
| 2579 | } |
| 2580 | if (handlers.empty()) { |
| 2581 | errorReporter.addError( |
| 2582 | kj::str("No event handlers were registered. This script does nothing.")); |
| 2583 | } |
| 2584 | errorReporter.addEntrypoint(kj::none, handlers.releaseAsArray()); |
| 2585 | } else { |
| 2586 | auto report = [&](kj::Maybe<kj::StringPtr> name, api::ExportedHandler& exported) { |
| 2587 | auto handle = exported.self.getHandle(js); |
| 2588 | if (handle->IsArray()) { |
| 2589 | // HACK: toDict() will throw a TypeError if given an array, because jsg::DictWrapper is |
| 2590 | // designed to treat arrays as not matching when a dict is expected. However, |
| 2591 | // StructWrapper has no such restriction, and therefore an exported array will |
| 2592 | // successfully produce an ExportedHandler (presumably with no handler functions), and |
| 2593 | // hence we will see it here. Rather than try to correct this inconsistency between |
| 2594 | // struct and dict handling (which could have unintended consequences), let's just |
| 2595 | // work around by ignoring arrays here. |
| 2596 | errorReporter.addEntrypoint(name, kj::Array<kj::String>()); |
| 2597 | } else { |
| 2598 | // Use a HashSet to avoid duplicates when methods exist both as own properties |
| 2599 | // and in the prototype chain |
| 2600 | kj::HashSet<kj::String> methodSet; |
| 2601 | |
| 2602 | // First, check for own properties (like a plain object literal) |
| 2603 | auto dict = js.toDict(handle); |
| 2604 | for (auto& field: dict.fields) { |
| 2605 | if (!ignoredHandlers.contains(field.name)) { |
| 2606 | methodSet.upsert(kj::mv(field.name), [&](auto&, auto&&) {}); |
| 2607 | } |
| 2608 | } |
| 2609 | |
| 2610 | // Then, check for methods in the prototype chain (like a class instance) |
| 2611 | js.withinHandleScope([&]() { |
| 2612 | collectMethodsFromPrototypeChain(jsg::JsObject(handle).getPrototype(js), methodSet); |
| 2613 | }); |
| 2614 | |
| 2615 | // Convert HashSet to Array for reporting |
| 2616 | errorReporter.addEntrypoint(name, KJ_MAP(n, methodSet) { return kj::mv(n); }); |
| 2617 | } |
| 2618 | }; |
| 2619 | |
| 2620 | auto getEntrypointName = [&](kj::StringPtr key) -> kj::Maybe<kj::StringPtr> { |
| 2621 | if (key == "default"_kj) { |
| 2622 | return kj::none; |
| 2623 | } else { |
| 2624 | return key; |
| 2625 | } |
| 2626 | }; |
| 2627 | |
| 2628 | for (auto& entry: worker.impl->namedHandlers) { |
| 2629 | report(getEntrypointName(entry.key), entry.value); |
| 2630 | } |
| 2631 | for (auto& entry: worker.impl->actorClasses) { |
| 2632 | KJ_IF_SOME(entrypointName, getEntrypointName(entry.key)) { |
| 2633 | errorReporter.addActorClass(entrypointName); |
| 2634 | } else { |
| 2635 | // Hmm, it appears someone tried to export a Durable Object class as a default |
| 2636 | // entrypoint. This doesn't actually work: the runtime will not allow this DO class |
| 2637 | // to be used, either for actors or as an entrypoint. |
| 2638 | // |
| 2639 | // TODO(someday): Make this a hard error. I'm hesitant to do it in my current change |
| 2640 | // for fear that it'll break someone somewhere forcing a rollback. For now we log. |
| 2641 | LOG_PERIODICALLY(ERROR, |
| 2642 | "Exported actor class as default entrypoint. This doesn't work, but historically " |
| 2643 | "did not produce a startup-time error."); |
| 2644 | } |
| 2645 | } |
| 2646 | for (auto& entry: worker.impl->statelessClasses) { |
| 2647 | // We want to report all of the stateless class's members. To do this, we examine its |
| 2648 | // prototype, and its prototype's prototype, and so on, until we get to Object's |
| 2649 | // prototype, which we ignore. |
| 2650 | auto entrypointName = getEntrypointName(entry.key); |
| 2651 | kj::HashSet<kj::String> seenNames; |
| 2652 | |
| 2653 | js.withinHandleScope([&]() { |
| 2654 | // For stateless classes, we need to get the class's prototype property |
| 2655 | jsg::JsObject ctor(KJ_ASSERT_NONNULL(entry.value.tryGetHandle(js.v8Isolate))); |
| 2656 | jsg::JsValue proto = ctor.get(js, "prototype"); |
| 2657 | collectMethodsFromPrototypeChain(proto, seenNames); |
| 2658 | }); |
| 2659 | |
| 2660 | errorReporter.addEntrypoint(entrypointName, KJ_MAP(n, seenNames) { return kj::mv(n); }); |
| 2661 | } |
| 2662 | |
| 2663 | for (auto& entry: worker.impl->workflowClasses) { |
| 2664 | KJ_IF_SOME(entrypointName, getEntrypointName(entry.key)) { |
| 2665 | kj::HashSet<kj::String> seenNames; |
| 2666 | |
| 2667 | js.withinHandleScope([&]() { |
| 2668 | // For stateless classes, we need to get the class's prototype property |
| 2669 | jsg::JsObject ctor(KJ_ASSERT_NONNULL(entry.value.tryGetHandle(js.v8Isolate))); |
| 2670 | jsg::JsValue proto = ctor.get(js, "prototype"); |
| 2671 | collectMethodsFromPrototypeChain(proto, seenNames); |
| 2672 | }); |
| 2673 | |
| 2674 | errorReporter.addWorkflowClass(entrypointName, KJ_MAP(n, seenNames) { return kj::mv(n); }); |
| 2675 | } else { |
| 2676 | } |
| 2677 | } |
| 2678 | } |
| 2679 | }); |
| 2680 | } |
| 2681 | |
| 2682 | // ======================================================================================= |
| 2683 | // AsyncLock implementation |
| 2684 | |
| 2685 | const kj::EventLoopLocal<Worker::AsyncWaiter*> Worker::AsyncWaiter::threadCurrentWaiter; |
| 2686 | |
| 2687 | Worker::Isolate::AsyncWaiterList::~AsyncWaiterList() noexcept { |
| 2688 | // It should be impossible for this list to be non-empty since each member of the list holds a |
| 2689 | // strong reference back to us. But if the list is non-empty, we'd better crash here, to avoid |
| 2690 | // dangling pointers. |
| 2691 | KJ_ASSERT(head == kj::none, "destroying non-empty waiter list?"); |
| 2692 | KJ_ASSERT(tail == &head, "tail pointer corrupted?"); |
| 2693 | } |
| 2694 | |
| 2695 | kj::Promise<Worker::AsyncLock> Worker::Isolate::takeAsyncLockWithoutRequest( |
| 2696 | SpanParent parentSpan) const { |
| 2697 | auto lockTiming = getMetrics().tryCreateLockTiming(kj::mv(parentSpan)); |
| 2698 | return takeAsyncLockImpl(kj::mv(lockTiming)); |
| 2699 | } |
| 2700 | |
| 2701 | kj::Promise<Worker::AsyncLock> Worker::Isolate::takeAsyncLock(RequestObserver& request) const { |
| 2702 | auto lockTiming = getMetrics().tryCreateLockTiming(kj::Maybe<RequestObserver&>(request)); |
| 2703 | return takeAsyncLockImpl(kj::mv(lockTiming)); |
| 2704 | } |
| 2705 | |
| 2706 | kj::Promise<Worker::AsyncLock> Worker::Isolate::takeAsyncLockImpl( |
| 2707 | kj::Maybe<kj::Own<IsolateObserver::LockTiming>> lockTiming) const { |
| 2708 | kj::Maybe<uint> currentLoad; |
| 2709 | if (lockTiming != kj::none) { |
| 2710 | currentLoad = getCurrentLoad(); |
| 2711 | } |
| 2712 | |
| 2713 | for (uint threadWaitingDifferentLockCount = 0;; ++threadWaitingDifferentLockCount) { |
| 2714 | AsyncWaiter* waiter = *AsyncWaiter::threadCurrentWaiter; |
| 2715 | |
| 2716 | if (waiter == nullptr) { |
| 2717 | // Thread is not currently waiting on a lock. |
| 2718 | KJ_IF_SOME(lt, lockTiming) { |
| 2719 | lt.get()->reportAsyncInfo(KJ_ASSERT_NONNULL(currentLoad), false /* threadWaitingSameLock */, |
| 2720 | threadWaitingDifferentLockCount); |
| 2721 | } |
| 2722 | auto newWaiter = kj::refcounted<AsyncWaiter>(kj::atomicAddRef(*this)); |
| 2723 | co_await newWaiter->readyPromise; |
| 2724 | co_return AsyncLock(kj::mv(newWaiter), kj::mv(lockTiming)); |
| 2725 | } else if (waiter->isolate == this) { |
| 2726 | // Thread is waiting on a lock already, and it's for the same isolate. We can coalesce the |
| 2727 | // locks. |
| 2728 | KJ_IF_SOME(lt, lockTiming) { |
| 2729 | lt.get()->reportAsyncInfo(KJ_ASSERT_NONNULL(currentLoad), true /* threadWaitingSameLock */, |
| 2730 | threadWaitingDifferentLockCount); |
| 2731 | } |
| 2732 | auto newWaiterRef = kj::addRef(*waiter); |
| 2733 | co_await newWaiterRef->readyPromise; |
| 2734 | co_return AsyncLock(kj::mv(newWaiterRef), kj::mv(lockTiming)); |
| 2735 | } else { |
| 2736 | // Thread is already waiting for or holding a different isolate lock. Wait for that one to |
| 2737 | // be released before we try to lock a different isolate. |
| 2738 | // TODO(perf): Use of ForkedPromise leads to thundering herd here. Should be minor in practice, |
| 2739 | // but we could consider creating another linked list instead... |
| 2740 | KJ_IF_SOME(lt, lockTiming) { |
| 2741 | lt.get()->waitingForOtherIsolate(waiter->isolate->getId()); |
| 2742 | } |
| 2743 | co_await waiter->releasePromise; |
| 2744 | } |
| 2745 | } |
| 2746 | } |
| 2747 | |
| 2748 | kj::Promise<Worker::AsyncLock> Worker::takeAsyncLockWithoutRequest(SpanParent parentSpan) const { |
| 2749 | return script->getIsolate().takeAsyncLockWithoutRequest(kj::mv(parentSpan)); |
| 2750 | } |
| 2751 | |
| 2752 | kj::Promise<Worker::AsyncLock> Worker::takeAsyncLock(RequestObserver& request) const { |
| 2753 | return script->getIsolate().takeAsyncLock(request); |
| 2754 | } |
| 2755 | |
| 2756 | Worker::AsyncWaiter::AsyncWaiter(kj::Own<const Isolate> isolateParam) |
| 2757 | : executor(kj::getCurrentThreadExecutor()), |
| 2758 | isolate(kj::mv(isolateParam)) { |
| 2759 | // Init `releasePromise` / `releaseFulfiller`. |
| 2760 | { |
| 2761 | auto paf = kj::newPromiseAndFulfiller<void>(); |
| 2762 | releasePromise = paf.promise.fork(); |
| 2763 | releaseFulfiller = kj::mv(paf.fulfiller); |
| 2764 | } |
| 2765 | |
| 2766 | // Add ourselves to the wait queue for this isolate. |
| 2767 | auto lock = isolate->asyncWaiters.lockExclusive(); |
| 2768 | if (lock->tail == &lock->head) { |
| 2769 | // Looks like the queue is empty, so we immediately get the lock. |
| 2770 | readyPromise = kj::Promise<void>(kj::READY_NOW).fork(); |
| 2771 | // We can leave `readyFulfiller` null as no one will ever invoke it anyway. |
| 2772 | } else { |
| 2773 | // Arrange to get notified later. |
| 2774 | auto paf = kj::newPromiseAndCrossThreadFulfiller<void>(); |
| 2775 | readyPromise = paf.promise.fork(); |
| 2776 | readyFulfiller = kj::mv(paf.fulfiller); |
| 2777 | } |
| 2778 | |
| 2779 | next = kj::none; |
| 2780 | prev = lock->tail; |
| 2781 | *lock->tail = this; |
| 2782 | lock->tail = &next; |
| 2783 | |
| 2784 | *threadCurrentWaiter = this; |
| 2785 | |
| 2786 | __atomic_add_fetch(&isolate->impl->lockAttemptGauge, 1, __ATOMIC_RELAXED); |
| 2787 | } |
| 2788 | |
| 2789 | Worker::AsyncWaiter::~AsyncWaiter() noexcept { |
| 2790 | // This destructor is `noexcept` because an exception here probably leaves the process in a bad |
| 2791 | // state. |
| 2792 | |
| 2793 | __atomic_sub_fetch(&isolate->impl->lockAttemptGauge, 1, __ATOMIC_RELAXED); |
| 2794 | |
| 2795 | auto lock = isolate->asyncWaiters.lockExclusive(); |
| 2796 | |
| 2797 | releaseFulfiller->fulfill(); |
| 2798 | |
| 2799 | // Remove ourselves from the list. |
| 2800 | *prev = next; |
| 2801 | KJ_IF_SOME(n, next) { |
| 2802 | n.prev = prev; |
| 2803 | } else { |
| 2804 | lock->tail = prev; |
| 2805 | } |
| 2806 | |
| 2807 | if (prev == &lock->head) { |
| 2808 | // We held the lock before now. Alert the next waiter that they are now at the front of the |
| 2809 | // line. |
| 2810 | KJ_IF_SOME(n, next) { |
| 2811 | n.readyFulfiller->fulfill(); |
| 2812 | } |
| 2813 | } |
| 2814 | |
| 2815 | auto& w = *threadCurrentWaiter; |
| 2816 | KJ_ASSERT(w == this); |
| 2817 | w = nullptr; |
| 2818 | } |
| 2819 | |
| 2820 | kj::Promise<void> Worker::AsyncLock::whenThreadIdle() { |
| 2821 | AsyncWaiter*& currentWaiter = *AsyncWaiter::threadCurrentWaiter; |
| 2822 | for (;;) { |
| 2823 | if (currentWaiter != nullptr) { |
| 2824 | co_await currentWaiter->releasePromise; |
| 2825 | continue; |
| 2826 | } |
| 2827 | |
| 2828 | co_await kj::yieldUntilQueueEmpty(); |
| 2829 | |
| 2830 | if (currentWaiter == nullptr) { |
| 2831 | co_return; |
| 2832 | } |
| 2833 | // Whoops, a new lock attempt appeared, loop. |
| 2834 | } |
| 2835 | } |
| 2836 | |
| 2837 | // ======================================================================================= |
| 2838 | |
| 2839 | // A proxy for OutputStream that internally buffers data as long as it's beyond a given limit. |
| 2840 | // Also, it counts size of all the data it has seen (whether it has hit the limit or not). |
| 2841 | // |
| 2842 | // We use this in the Network tab to report response stats and preview [decompressed] bodies, |
| 2843 | // but we don't want to keep buffering extremely large ones, so just discard buffered data |
| 2844 | // upon hitting a limit and don't return any body to the devtools frontend afterwards. |
| 2845 | class Worker::Isolate::LimitedBodyWrapper: public kj::OutputStream { |
| 2846 | public: |
| 2847 | LimitedBodyWrapper(size_t limit = 1 * 1024 * 1024): limit(limit) { |
| 2848 | if (limit > 0) { |
| 2849 | inner.emplace(); |
| 2850 | } |
| 2851 | } |
| 2852 | |
| 2853 | KJ_DISALLOW_COPY_AND_MOVE(LimitedBodyWrapper); |
| 2854 | |
| 2855 | void reset() { |
| 2856 | this->inner = kj::none; |
| 2857 | } |
| 2858 | |
| 2859 | void write(kj::ArrayPtr<const byte> data) override { |
| 2860 | this->size += data.size(); |
| 2861 | KJ_IF_SOME(inner, this->inner) { |
| 2862 | if (this->size <= this->limit) { |
| 2863 | inner.write(data); |
| 2864 | } else { |
| 2865 | reset(); |
| 2866 | } |
| 2867 | } |
| 2868 | } |
| 2869 | |
| 2870 | size_t getWrittenSize() { |
| 2871 | return this->size; |
| 2872 | } |
| 2873 | |
| 2874 | kj::Maybe<kj::ArrayPtr<byte>> getArray() { |
| 2875 | KJ_IF_SOME(inner, this->inner) { |
| 2876 | return inner.getArray(); |
| 2877 | } else { |
| 2878 | return kj::none; |
| 2879 | } |
| 2880 | } |
| 2881 | |
| 2882 | private: |
| 2883 | size_t size = 0; |
| 2884 | size_t limit = 0; |
| 2885 | kj::Maybe<kj::VectorOutputStream> inner; |
| 2886 | }; |
| 2887 | |
| 2888 | struct MessageQueue { |
| 2889 | kj::Vector<kj::String> messages; |
| 2890 | size_t head; |
| 2891 | enum class Status { ACTIVE, CLOSED } status; |
| 2892 | }; |
| 2893 | |
| 2894 | class Worker::Isolate::InspectorChannelImpl final: public v8_inspector::V8Inspector::Channel { |
| 2895 | public: |
| 2896 | InspectorChannelImpl(kj::Own<const Worker::Isolate> isolateParam, |
| 2897 | kj::Own<const kj::Executor> isolateThreadExecutor, |
| 2898 | kj::WebSocket& webSocket) |
| 2899 | : ioHandler(kj::mv(isolateThreadExecutor), webSocket), |
| 2900 | state(kj::heap<State>(this, kj::mv(isolateParam))) { |
| 2901 | ioHandler.connect(*this); |
| 2902 | } |
| 2903 | |
| 2904 | // In preview sessions, synchronous locks are not an issue. We declare an alternate spelling of |
| 2905 | // the type so that all the individual locks below don't turn up in a search for synchronous |
| 2906 | // locks. |
| 2907 | using InspectorLock = Worker::Lock::TakeSynchronously; |
| 2908 | |
| 2909 | ~InspectorChannelImpl() noexcept try { |
| 2910 | // Stop message pump. |
| 2911 | ioHandler.disconnect(); |
| 2912 | |
| 2913 | // Delete session under lock. |
| 2914 | auto state = this->state.lockExclusive(); |
| 2915 | |
| 2916 | jsg::runInV8Stack([&](jsg::V8StackScope& stackScope) { |
| 2917 | Isolate::Impl::Lock recordedLock(*state->get()->isolate, InspectorLock(kj::none), stackScope); |
| 2918 | if (state->get()->isolate->currentInspectorSession != kj::none) { |
| 2919 | const_cast<Isolate&>(*state->get()->isolate).disconnectInspector(); |
| 2920 | } |
| 2921 | state->get()->teardownUnderLock(); |
| 2922 | }); |
| 2923 | } catch (...) { |
| 2924 | // Unfortunately since we're inheriting from Channel which declares a virtual destructor with |
| 2925 | // default exception constraints, we have to catch all exceptions here and log them. |
| 2926 | // But different kinds of exceptions call for different ways to stringify the exception. |
| 2927 | // kj::runCatchingExceptions() normally does this for us, but there's no way to use it while |
| 2928 | // wrapping the whole destructor (including destructors of members). So... we do a native |
| 2929 | // catch(...) and then we rethrow the exception inside a kj::runCatchingExceptions and then log |
| 2930 | // that. Yeah. |
| 2931 | // |
| 2932 | // TODO(cleanup): Maybe we could add a kj::stringifyCurrentException() or |
| 2933 | // kj::logUncaughtException() or something? |
| 2934 | KJ_IF_SOME(exception, kj::runCatchingExceptions([&]() { throw; })) { |
| 2935 | KJ_LOG(ERROR, "uncaught exception in ~Script() and the C++ standard is broken", exception); |
| 2936 | } |
| 2937 | } |
| 2938 | |
| 2939 | void disconnect() { |
| 2940 | // Fake like the client requested close. This will cause outgoingLoop() to exit and everything |
| 2941 | // will be cleaned up. |
| 2942 | ioHandler.disconnect(); |
| 2943 | } |
| 2944 | |
| 2945 | void dispatchProtocolMessage(kj::String message, |
| 2946 | v8_inspector::V8InspectorSession& session, |
| 2947 | Isolate& isolate, |
| 2948 | jsg::V8StackScope& stackScope, |
| 2949 | Isolate::Impl::Lock& recordedLock) { |
| 2950 | capnp::MallocMessageBuilder messageBuilder; |
| 2951 | auto cmd = messageBuilder.initRoot<cdp::Command>(); |
| 2952 | getCdpJsonCodec().decode(message, cmd); |
| 2953 | |
| 2954 | switch (cmd.which()) { |
| 2955 | case cdp::Command::UNKNOWN: { |
| 2956 | break; |
| 2957 | } |
| 2958 | case cdp::Command::NETWORK_ENABLE: { |
| 2959 | setNetworkEnabled(true); |
| 2960 | cmd.getNetworkEnable().initResult(); |
| 2961 | break; |
| 2962 | } |
| 2963 | case cdp::Command::NETWORK_DISABLE: { |
| 2964 | setNetworkEnabled(false); |
| 2965 | cmd.getNetworkDisable().initResult(); |
| 2966 | break; |
| 2967 | } |
| 2968 | case cdp::Command::NETWORK_GET_RESPONSE_BODY: { |
| 2969 | auto err = cmd.getNetworkGetResponseBody().initError(); |
| 2970 | err.setCode(-32600); |
| 2971 | err.setMessage("Network.getResponseBody is not supported in this fork"); |
| 2972 | break; |
| 2973 | } |
| 2974 | case cdp::Command::PROFILER_STOP: { |
| 2975 | KJ_IF_SOME(p, isolate.impl->profiler) { |
| 2976 | auto& lock = recordedLock.lock; |
| 2977 | stopProfiling(*lock, *p, cmd); |
| 2978 | } |
| 2979 | break; |
| 2980 | } |
| 2981 | case cdp::Command::PROFILER_START: { |
| 2982 | KJ_IF_SOME(p, isolate.impl->profiler) { |
| 2983 | auto& lock = recordedLock.lock; |
| 2984 | startProfiling(*lock, *p); |
| 2985 | } |
| 2986 | break; |
| 2987 | } |
| 2988 | case cdp::Command::PROFILER_SET_SAMPLING_INTERVAL: { |
| 2989 | KJ_IF_SOME(p, isolate.impl->profiler) { |
| 2990 | auto interval = cmd.getProfilerSetSamplingInterval().getParams().getInterval(); |
| 2991 | setSamplingInterval(*p, interval); |
| 2992 | } |
| 2993 | break; |
| 2994 | } |
| 2995 | case cdp::Command::PROFILER_ENABLE: { |
| 2996 | auto& lock = recordedLock.lock; |
| 2997 | isolate.impl->profiler = kj::Own<v8::CpuProfiler>( |
| 2998 | v8::CpuProfiler::New(lock->v8Isolate, v8::kDebugNaming, v8::kLazyLogging), |
| 2999 | CpuProfilerDisposer::instance); |
| 3000 | break; |
| 3001 | } |
| 3002 | case cdp::Command::TAKE_HEAP_SNAPSHOT: { |
| 3003 | auto& lock = recordedLock.lock; |
| 3004 | takeHeapSnapshot(*lock, cmd.getTakeHeapSnapshot().getParams()); |
| 3005 | break; |
| 3006 | } |
| 3007 | } |
| 3008 | |
| 3009 | if (!cmd.isUnknown()) { |
| 3010 | sendNotification(cmd); |
| 3011 | return; |
| 3012 | } |
| 3013 | |
| 3014 | auto& lock = recordedLock.lock; |
| 3015 | |
| 3016 | // We have at times observed V8 bugs where the inspector queues a background task and |
| 3017 | // then synchronously waits for it to complete, which would deadlock if background |
| 3018 | // threads are disallowed. Since the inspector is in a process sandbox anyway, it's not |
| 3019 | // a big deal to just permit those background threads. |
| 3020 | AllowV8BackgroundThreadsScope allowBackgroundThreads; |
| 3021 | |
| 3022 | ExceptionOrDuration limitErrorOrTime = 0 * kj::NANOSECONDS; |
| 3023 | { |
| 3024 | auto limitScope = isolate.getLimitEnforcer().enterInspectorJs(*lock, limitErrorOrTime); |
| 3025 | session.dispatchProtocolMessage(jsg::toInspectorStringView(message)); |
| 3026 | } |
| 3027 | |
| 3028 | // Run microtasks in case the user made an async call. |
| 3029 | if (!limitErrorOrTime.is<kj::Exception>()) { |
| 3030 | auto limitScope = isolate.getLimitEnforcer().enterInspectorJs(*lock, limitErrorOrTime); |
| 3031 | lock->runMicrotasks(); |
| 3032 | } else { |
| 3033 | // Oops, we already exceeded the limit, so force the microtask queue to be thrown away. |
| 3034 | lock->terminateNextExecution(); |
| 3035 | lock->runMicrotasks(); |
| 3036 | } |
| 3037 | |
| 3038 | KJ_SWITCH_ONEOF(limitErrorOrTime) { |
| 3039 | KJ_CASE_ONEOF(limitError, kj::Exception) { |
| 3040 | lock->withinHandleScope([&] { |
| 3041 | // HACK: We want to print the error, but we need a context to do that. |
| 3042 | // We don't know which contexts exist in this isolate, so I guess we have to |
| 3043 | // create one. Ugh. |
| 3044 | auto dummyContext = v8::Context::New(lock->v8Isolate); |
| 3045 | // We need to set the highest used index in every context we create to be a nullptr |
| 3046 | // This is because we might later on call GetAlignedPointerFromEmbedderData which fails with |
| 3047 | // a fatal error if the array is smaller than the given index. |
| 3048 | jsg::setAlignedPointerInEmbedderData( |
| 3049 | dummyContext, jsg::ContextPointerSlot::MAX_POINTER_SLOT, nullptr); |
| 3050 | auto& inspector = *KJ_ASSERT_NONNULL(isolate.impl->inspector); |
| 3051 | inspector.contextCreated(v8_inspector::V8ContextInfo(dummyContext, 1, |
| 3052 | v8_inspector::StringView(reinterpret_cast<const uint8_t*>("Worker"), 6))); |
| 3053 | JSG_WITHIN_CONTEXT_SCOPE(*lock, dummyContext, [&](jsg::Lock& js) { |
| 3054 | jsg::sendExceptionToInspector(js, inspector, |
| 3055 | jsg::extractTunneledExceptionDescription(limitError.getDescription())); |
| 3056 | }); |
| 3057 | inspector.contextDestroyed(dummyContext); |
| 3058 | }); |
| 3059 | } |
| 3060 | KJ_CASE_ONEOF(startupTimeElapsed, kj::Duration) {} |
| 3061 | } |
| 3062 | |
| 3063 | if (recordedLock.checkInWithLimitEnforcer(isolate)) { |
| 3064 | disconnect(); |
| 3065 | } |
| 3066 | } |
| 3067 | |
| 3068 | kj::Promise<void> messagePump() { |
| 3069 | return ioHandler.messagePump(); |
| 3070 | } |
| 3071 | |
| 3072 | void handleDispatchProtocolMessage( |
| 3073 | Worker::AsyncLock& asyncLock, kj::MutexGuarded<MessageQueue>& incomingQueue) { |
| 3074 | auto lockedState = state.lockExclusive(); |
| 3075 | v8_inspector::V8InspectorSession& session = *lockedState->get()->session; |
| 3076 | Isolate& isolate = const_cast<Isolate&>(*lockedState->get()->isolate); |
| 3077 | jsg::runInV8Stack([&](jsg::V8StackScope& stackScope) { |
| 3078 | Isolate::Impl::Lock recordedLock(isolate, asyncLock, stackScope); |
| 3079 | |
| 3080 | auto lockedQueue = incomingQueue.lockExclusive(); |
| 3081 | if (lockedQueue->status != MessageQueue::Status::ACTIVE) { |
| 3082 | return; |
| 3083 | } |
| 3084 | |
| 3085 | auto messages = lockedQueue->messages.slice(lockedQueue->head, lockedQueue->messages.size()); |
| 3086 | for (auto& message: messages) { |
| 3087 | dispatchProtocolMessage(kj::mv(message), session, isolate, stackScope, recordedLock); |
| 3088 | } |
| 3089 | lockedQueue->messages.clear(); |
| 3090 | lockedQueue->head = 0; |
| 3091 | }); |
| 3092 | } |
| 3093 | |
| 3094 | kj::Promise<void> dispatchProtocolMessages(kj::MutexGuarded<MessageQueue>& incomingQueue) { |
| 3095 | // This method is called on the I/O thread, which also adds messages to the `incomingQueue`. |
| 3096 | // So long as this method does not yield/resume mid-way, there is no concern about how |
| 3097 | // long the queue lock is held for whilst dispatching messages. |
| 3098 | auto i = kj::atomicAddRef(*this->state.lockExclusive()->get()->isolate); |
| 3099 | auto asyncLock = co_await i->takeAsyncLockWithoutRequest(nullptr); |
| 3100 | handleDispatchProtocolMessage(asyncLock, incomingQueue); |
| 3101 | } |
| 3102 | |
| 3103 | // --------------------------------------------------------------------------- |
| 3104 | // implements Channel |
| 3105 | // |
| 3106 | // Keep in mind that these methods will be called from various threads! |
| 3107 | |
| 3108 | void sendResponse(int callId, std::unique_ptr<v8_inspector::StringBuffer> message) override { |
| 3109 | // callId is encoded in the message, too. Unsure why this method even exists. |
| 3110 | sendNotification(kj::mv(message)); |
| 3111 | } |
| 3112 | |
| 3113 | bool isNetworkEnabled() { |
| 3114 | return __atomic_load_n(&networkEnabled, __ATOMIC_RELAXED); |
| 3115 | } |
| 3116 | |
| 3117 | void setNetworkEnabled(bool enable) { |
| 3118 | __atomic_store_n(&networkEnabled, enable, __ATOMIC_RELAXED); |
| 3119 | } |
| 3120 | |
| 3121 | void sendNotification(kj::String message) { |
| 3122 | ioHandler.send(kj::mv(message)); |
| 3123 | } |
| 3124 | |
| 3125 | template <typename T> |
| 3126 | void sendNotification(T&& message) { |
| 3127 | sendNotification(getCdpJsonCodec().encode(message)); |
| 3128 | } |
| 3129 | |
| 3130 | void sendNotification(std::unique_ptr<v8_inspector::StringBuffer> message) override { |
| 3131 | sendNotification(kj::str(message->string())); |
| 3132 | } |
| 3133 | |
| 3134 | void flushProtocolNotifications() override { |
| 3135 | // Are we supposed to do anything here? There's no documentation, so who knows? Maybe we could |
| 3136 | // delay signaling the outgoing loop until this call? |
| 3137 | } |
| 3138 | |
| 3139 | // Dispatches one message whilst automatic CDP messages on the I/O worker thread is paused, called |
| 3140 | // on the thread executing the isolate whilst execution is suspended due to a breakpoint or |
| 3141 | // debugger statement. |
| 3142 | bool dispatchOneMessageDuringPause(); |
| 3143 | |
| 3144 | private: |
| 3145 | // Class that manages the I/O for devtools connections. I/O is performed on the |
| 3146 | // thread associated with the InspectorService (the thread that calls attachInspector). |
| 3147 | // Most of the public API is intended for code running on the isolate thread, such as |
| 3148 | // the InspectorChannelImpl and the InspectorClient. |
| 3149 | class WebSocketIoHandler final { |
| 3150 | public: |
| 3151 | WebSocketIoHandler(kj::Own<const kj::Executor> isolateThreadExecutor, kj::WebSocket& webSocket) |
| 3152 | : isolateThreadExecutor(kj::mv(isolateThreadExecutor)), |
| 3153 | webSocket(webSocket) { |
| 3154 | // Assume we are being instantiated on the InspectorService thread, the thread that will do |
| 3155 | // I/O for CDP messages. Messages are delivered to the InspectorChannelImpl on the Isolate thread. |
| 3156 | outgoingQueueNotifier = XThreadNotifier::create(); |
| 3157 | } |
| 3158 | |
| 3159 | // Sets the channel that messages are delivered to. |
| 3160 | void connect(InspectorChannelImpl& inspectorChannel) { |
| 3161 | channel = inspectorChannel; |
| 3162 | } |
| 3163 | |
| 3164 | void disconnect() { |
| 3165 | channel = kj::none; |
| 3166 | shutdown(); |
| 3167 | } |
| 3168 | |
| 3169 | // Blocked the current thread until a message arrives. This is intended |
| 3170 | // for use in the InspectorClient when breakpoints are hit. The InspectorClient |
| 3171 | // has to remain in runMessageLoopOnPause() but still receive CDP messages |
| 3172 | // (e.g. resume). |
| 3173 | kj::Maybe<kj::String> waitForMessage() { |
| 3174 | return incomingQueue.when([](const MessageQueue& incomingQueue) { |
| 3175 | return (incomingQueue.head < incomingQueue.messages.size() || |
| 3176 | incomingQueue.status == MessageQueue::Status::CLOSED); |
| 3177 | }, [](MessageQueue& incomingQueue) -> kj::Maybe<kj::String> { |
| 3178 | if (incomingQueue.status == MessageQueue::Status::CLOSED) return {}; |
| 3179 | return pollMessage(incomingQueue); |
| 3180 | }); |
| 3181 | } |
| 3182 | |
| 3183 | // Message pumping promise that should be evaluated on the InspectorService |
| 3184 | // thread. |
| 3185 | kj::Promise<void> messagePump() { |
| 3186 | // Although inspector I/O must happen on the InspectorService thread (to make sure breakpoints |
| 3187 | // don't block inspector I/O), inspector messages must be actually dispatched on the Isolate |
| 3188 | // thread. So, we run the dispatch loop on the Isolate thread. |
| 3189 | // |
| 3190 | // Note that the above comment is only really accurate in vanilla workerd. In the case of the |
| 3191 | // internal Cloudflare Workers runtime, `isolateThreadExecutor` may actually refer to the |
| 3192 | // current thread's `kj::Executor`. That's fine; calling `executeAsync()` on the current |
| 3193 | // thread's executor just posts the task to the event loop, and everything works as expected. |
| 3194 | |
| 3195 | // Since the dispatch loop and the receive loop communicate over a XThreadNotifier, and |
| 3196 | // XThreadNotifiers must be created on the thread which will call their `awaitNotification()` |
| 3197 | // function, we awkwardly perform two `executeAsync()`s here, one to create the |
| 3198 | // XThreadNotifier, then another to spawn the dispatch loop. |
| 3199 | // |
| 3200 | // We create a new XThreadNotifier for each `messagePump()` call, rather than try to re-use |
| 3201 | // one long-term, because XThreadNotifiers' `awaitNotification()` function is not cancel-safe. |
| 3202 | // That is, once its promise is cancelled, the notifier is broken. |
| 3203 | auto incomingQueueNotifier = |
| 3204 | co_await isolateThreadExecutor->executeAsync([]() { return XThreadNotifier::create(); }); |
| 3205 | |
| 3206 | auto dispatchLoopPromise = isolateThreadExecutor->executeAsync( |
| 3207 | [this, notifier = kj::atomicAddRef(*incomingQueueNotifier)]() mutable { |
| 3208 | return dispatchLoop(kj::mv(notifier)); |
| 3209 | }); |
| 3210 | |
| 3211 | co_return co_await receiveLoop(kj::mv(incomingQueueNotifier)) |
| 3212 | .exclusiveJoin(kj::mv(dispatchLoopPromise)) |
| 3213 | .exclusiveJoin(transmitLoop()); |
| 3214 | } |
| 3215 | |
| 3216 | void send(kj::String message) { |
| 3217 | auto lockedOutgoingQueue = outgoingQueue.lockExclusive(); |
| 3218 | if (lockedOutgoingQueue->status == MessageQueue::Status::CLOSED) return; |
| 3219 | lockedOutgoingQueue->messages.add(kj::mv(message)); |
| 3220 | outgoingQueueNotifier->notify(); |
| 3221 | } |
| 3222 | |
| 3223 | private: |
| 3224 | static kj::Maybe<kj::String> pollMessage(MessageQueue& messageQueue) { |
| 3225 | if (messageQueue.head < messageQueue.messages.size()) { |
| 3226 | kj::String message = kj::mv(messageQueue.messages[messageQueue.head++]); |
| 3227 | if (messageQueue.head == messageQueue.messages.size()) { |
| 3228 | messageQueue.head = 0; |
| 3229 | messageQueue.messages.clear(); |
| 3230 | } |
| 3231 | return kj::mv(message); |
| 3232 | } |
| 3233 | return {}; |
| 3234 | } |
| 3235 | |
| 3236 | void shutdown() { |
| 3237 | // Drain incoming queue, the isolate thread may be waiting on it |
| 3238 | // on will notice it is closed if woken without any messages to |
| 3239 | // deliver in WebSocketIoWorker::waitForMessage(). |
| 3240 | { |
| 3241 | auto lockedIncomingQueue = incomingQueue.lockExclusive(); |
| 3242 | lockedIncomingQueue->head = 0; |
| 3243 | lockedIncomingQueue->messages.clear(); |
| 3244 | lockedIncomingQueue->status = MessageQueue::Status::CLOSED; |
| 3245 | } |
| 3246 | { |
| 3247 | auto lockedOutgoingQueue = outgoingQueue.lockExclusive(); |
| 3248 | lockedOutgoingQueue->status = MessageQueue::Status::CLOSED; |
| 3249 | } |
| 3250 | // Wake any waiters since queue status fields have been updated. |
| 3251 | outgoingQueueNotifier->notify(); |
| 3252 | } |
| 3253 | |
| 3254 | // Must be called on the InspectorService thread. |
| 3255 | kj::Promise<void> receiveLoop(kj::Own<XThreadNotifier> incomingQueueNotifier) { |
| 3256 | for (;;) { |
| 3257 | auto message = co_await webSocket.receive(MAX_MESSAGE_SIZE); |
| 3258 | KJ_SWITCH_ONEOF(message) { |
| 3259 | KJ_CASE_ONEOF(text, kj::String) { |
| 3260 | incomingQueue.lockExclusive()->messages.add(kj::mv(text)); |
| 3261 | incomingQueueNotifier->notify(); |
| 3262 | } |
| 3263 | KJ_CASE_ONEOF(blob, kj::Array<byte>) { |
| 3264 | // Ignore. |
| 3265 | } |
| 3266 | KJ_CASE_ONEOF(close, kj::WebSocket::Close) { |
| 3267 | shutdown(); |
| 3268 | // Pause here to give transmitLoop() the chance to finish and send a reply close. |
| 3269 | // When `transmitLoop()` ends, `messagePump()` as a whole will end, canceling |
| 3270 | // `receiveLoop()`. |
| 3271 | co_await kj::Promise<void>(kj::NEVER_DONE); |
| 3272 | } |
| 3273 | } |
| 3274 | } |
| 3275 | } |
| 3276 | |
| 3277 | // Must be called on the Isolate thread. |
| 3278 | kj::Promise<void> dispatchLoop(kj::Own<XThreadNotifier> incomingQueueNotifier) { |
| 3279 | for (;;) { |
| 3280 | co_await incomingQueueNotifier->awaitNotification(); |
| 3281 | KJ_IF_SOME(c, channel) { |
| 3282 | co_await c.dispatchProtocolMessages(this->incomingQueue); |
| 3283 | } |
| 3284 | } |
| 3285 | } |
| 3286 | |
| 3287 | // Must be called on the InspectorService thread. |
| 3288 | kj::Promise<void> transmitLoop() { |
| 3289 | for (;;) { |
| 3290 | co_await outgoingQueueNotifier->awaitNotification(); |
| 3291 | try { |
| 3292 | auto lockedOutgoingQueue = outgoingQueue.lockExclusive(); |
| 3293 | auto messages = kj::mv(lockedOutgoingQueue->messages); |
| 3294 | bool receivedClose = lockedOutgoingQueue->status == MessageQueue::Status::CLOSED; |
| 3295 | lockedOutgoingQueue.release(); |
| 3296 | co_await sendToWebSocket(kj::mv(messages)); |
| 3297 | if (receivedClose) { |
| 3298 | co_await webSocket.close(1000, "client closed connection"); |
| 3299 | co_return; |
| 3300 | } |
| 3301 | } catch (kj::Exception& e) { |
| 3302 | shutdown(); |
| 3303 | throw; |
| 3304 | } |
| 3305 | } |
| 3306 | } |
| 3307 | |
| 3308 | kj::Promise<void> sendToWebSocket(kj::Vector<kj::String> messages) { |
| 3309 | for (auto& message: messages) { |
| 3310 | co_await webSocket.send(message); |
| 3311 | } |
| 3312 | } |
| 3313 | |
| 3314 | // We need access to the Isolate thread's kj::Executor to run the inspector dispatch loop. This |
| 3315 | // doesn't actually have to be an Own, because the Isolate thread will destroy the Isolate |
| 3316 | // before it exits, but it doesn't hurt. |
| 3317 | kj::Own<const kj::Executor> isolateThreadExecutor; |
| 3318 | |
| 3319 | kj::MutexGuarded<MessageQueue> incomingQueue; |
| 3320 | // The notifier for `incomingQueue`, `incomingQueueNotifier`, is created once per |
| 3321 | // `messagePump()` call, and never re-used, so it doesn't live here. |
| 3322 | |
| 3323 | kj::MutexGuarded<MessageQueue> outgoingQueue; |
| 3324 | // This XThreadNotifier must be created on the InspectorService thread. |
| 3325 | kj::Own<XThreadNotifier> outgoingQueueNotifier; |
| 3326 | |
| 3327 | kj::WebSocket& webSocket; // only accessed on the InspectorService thread. |
| 3328 | std::atomic_bool receivedClose; // accessed on any thread (only transitions false -> true). |
| 3329 | kj::Maybe<InspectorChannelImpl&> channel; // only accessed on the isolate thread. |
| 3330 | |
| 3331 | // Sometimes the inspector protocol sends large messages. KJ defaults to a 1MB size limit |
| 3332 | // for WebSocket messages, which makes sense for production use cases, but for debug we should |
| 3333 | // be OK to go larger. So, we'll accept 128MB. |
| 3334 | static constexpr size_t MAX_MESSAGE_SIZE = 128u << 20; |
| 3335 | }; |
| 3336 | |
| 3337 | WebSocketIoHandler ioHandler; |
| 3338 | |
| 3339 | void takeHeapSnapshot( |
| 3340 | jsg::Lock& js, cdp::HeapProfiler::Command::TakeHeapSnapshot::Params::Reader params) { |
| 3341 | struct Activity: public v8::ActivityControl { |
| 3342 | InspectorChannelImpl& channel; |
| 3343 | Activity(InspectorChannelImpl& channel): channel(channel) {} |
| 3344 | |
| 3345 | ControlOption ReportProgressValue(uint32_t done, uint32_t total) override { |
| 3346 | capnp::MallocMessageBuilder message; |
| 3347 | auto event = message.initRoot<cdp::Event>(); |
| 3348 | auto progressParams = event.initReportHeapSnapshotProgress(); |
| 3349 | progressParams.setDone(done); |
| 3350 | progressParams.setTotal(total); |
| 3351 | if (done == total) { |
| 3352 | progressParams.setFinished(true); |
| 3353 | } |
| 3354 | auto notification = getCdpJsonCodec().encode(event); |
| 3355 | channel.sendNotification(kj::mv(notification)); |
| 3356 | return ControlOption::kContinue; |
| 3357 | } |
| 3358 | }; |
| 3359 | |
| 3360 | struct Writer: public v8::OutputStream { |
| 3361 | InspectorChannelImpl& channel; |
| 3362 | |
| 3363 | Writer(InspectorChannelImpl& channel): channel(channel) {} |
| 3364 | void EndOfStream() override {} |
| 3365 | |
| 3366 | int GetChunkSize() override { |
| 3367 | return 65536; // big chunks == faster |
| 3368 | // The chunk size here will determine the actual number of individual |
| 3369 | // messages that are sent. The default is... rather small. Experience |
| 3370 | // node and node-heapdump shows that this can be bumped up |
| 3371 | // much higher to get better performance. Here we use the value |
| 3372 | // that Node.js uses (see Node.js' FileOutputStream impl). |
| 3373 | } |
| 3374 | |
| 3375 | v8::OutputStream::WriteResult WriteAsciiChunk(char* data, int size) override { |
| 3376 | capnp::MallocMessageBuilder message; |
| 3377 | auto event = message.initRoot<cdp::Event>(); |
| 3378 | |
| 3379 | auto params = event.initAddHeapSnapshotChunk(); |
| 3380 | params.setChunk(kj::heapString(data, size)); |
| 3381 | auto notification = getCdpJsonCodec().encode(event); |
| 3382 | channel.sendNotification(kj::mv(notification)); |
| 3383 | |
| 3384 | return v8::OutputStream::WriteResult::kContinue; |
| 3385 | } |
| 3386 | }; |
| 3387 | |
| 3388 | Activity activity(*this); |
| 3389 | Writer writer(*this); |
| 3390 | |
| 3391 | v8::HeapProfiler::HeapSnapshotOptions options{}; |
| 3392 | if (params.getReportProgress()) { |
| 3393 | options.control = &activity; |
| 3394 | } |
| 3395 | if (params.getExposeInternals()) { |
| 3396 | options.snapshot_mode = v8::HeapProfiler::HeapSnapshotMode::kExposeInternals; |
| 3397 | } |
| 3398 | if (params.getCaptureNumericValue()) { |
| 3399 | options.numerics_mode = v8::HeapProfiler::NumericsMode::kExposeNumericValues; |
| 3400 | } |
| 3401 | |
| 3402 | auto profiler = js.v8Isolate->GetHeapProfiler(); |
| 3403 | auto snapshot = kj::Own<const v8::HeapSnapshot>( |
| 3404 | profiler->TakeHeapSnapshot(options), HeapSnapshotDeleter::INSTANCE); |
| 3405 | snapshot->Serialize(&writer); |
| 3406 | } |
| 3407 | |
| 3408 | struct State { |
| 3409 | kj::Own<const Worker::Isolate> isolate; |
| 3410 | std::unique_ptr<v8_inspector::V8InspectorSession> session; |
| 3411 | |
| 3412 | State(InspectorChannelImpl* self, kj::Own<const Worker::Isolate> isolateParam) |
| 3413 | : isolate(kj::mv(isolateParam)), |
| 3414 | session(KJ_ASSERT_NONNULL(isolate->impl->inspector) |
| 3415 | ->connect(1, |
| 3416 | self, |
| 3417 | v8_inspector::StringView(), |
| 3418 | isolate->impl->inspectorPolicy == InspectorPolicy::ALLOW_UNTRUSTED |
| 3419 | ? v8_inspector::V8Inspector::kUntrusted |
| 3420 | : v8_inspector::V8Inspector::kFullyTrusted)) {} |
| 3421 | ~State() noexcept(false) { |
| 3422 | if (session != nullptr) { |
| 3423 | KJ_LOG(ERROR, |
| 3424 | "Deleting InspectorChannelImpl::State without having called " |
| 3425 | "teardownUnderLock()", |
| 3426 | kj::getStackTrace()); |
| 3427 | |
| 3428 | // Isolate locks are recursive so it should be safe to lock here. |
| 3429 | jsg::runInV8Stack([&](jsg::V8StackScope& stackScope) { |
| 3430 | Isolate::Impl::Lock recordedLock(*isolate, InspectorLock(kj::none), stackScope); |
| 3431 | session = nullptr; |
| 3432 | }); |
| 3433 | } |
| 3434 | } |
| 3435 | |
| 3436 | // Must be called with the worker isolate locked. Should be called immediately before |
| 3437 | // destruction. |
| 3438 | void teardownUnderLock() { |
| 3439 | session = nullptr; |
| 3440 | } |
| 3441 | |
| 3442 | KJ_DISALLOW_COPY_AND_MOVE(State); |
| 3443 | }; |
| 3444 | // Mutex ordering: You must lock this *before* locking the isolate. |
| 3445 | kj::MutexGuarded<kj::Own<State>> state; |
| 3446 | |
| 3447 | // Not under `state` lock due to lock ordering complications. |
| 3448 | volatile bool networkEnabled = false; |
| 3449 | }; |
| 3450 | |
| 3451 | bool Worker::Isolate::InspectorChannelImpl::dispatchOneMessageDuringPause() { |
| 3452 | auto maybeMessage = ioHandler.waitForMessage(); |
| 3453 | // We can be paused by either hitting a debugger statement in a script or from hitting |
| 3454 | // a breakpoint or someone hit break. |
| 3455 | KJ_IF_SOME(message, maybeMessage) { |
| 3456 | auto lockedState = this->state.lockExclusive(); |
| 3457 | // Received a message whilst script is running, probably in a breakpoint. |
| 3458 | v8_inspector::V8InspectorSession& session = *lockedState->get()->session; |
| 3459 | // const_cast OK because the IoContext has the lock. |
| 3460 | Isolate& isolate = const_cast<Isolate&>(*lockedState->get()->isolate); |
| 3461 | Worker::Lock& workerLock = IoContext::current().getCurrentLock(); |
| 3462 | Isolate::Impl::Lock& recordedLock = workerLock.impl->recordedLock; |
| 3463 | jsg::runInV8Stack([&](jsg::V8StackScope& stackScope) { |
| 3464 | dispatchProtocolMessage(kj::mv(message), session, isolate, stackScope, recordedLock); |
| 3465 | }); |
| 3466 | return true; |
| 3467 | } else { |
| 3468 | // No message from waitForMessage() implies the connection is broken. |
| 3469 | return false; |
| 3470 | } |
| 3471 | } |
| 3472 | |
| 3473 | bool Worker::InspectorClient::dispatchOneMessageDuringPause( |
| 3474 | Worker::Isolate::InspectorChannelImpl& channel) { |
| 3475 | return channel.dispatchOneMessageDuringPause(); |
| 3476 | } |
| 3477 | |
| 3478 | kj::Promise<void> Worker::Isolate::attachInspector(kj::Timer& timer, |
| 3479 | kj::Duration timerOffset, |
| 3480 | kj::HttpService::Response& response, |
| 3481 | const kj::HttpHeaderTable& headerTable, |
| 3482 | kj::HttpHeaderId controlHeaderId) const { |
| 3483 | KJ_REQUIRE(impl->inspector != kj::none); |
| 3484 | |
| 3485 | kj::HttpHeaders headers(headerTable); |
| 3486 | headers.setPtr(controlHeaderId, "{\"ewLog\":{\"status\":\"ok\"}}"); |
| 3487 | auto webSocket = response.acceptWebSocket(headers); |
| 3488 | |
| 3489 | // This `attachInspector()` overload is used by the internal Cloudflare Workers runtime, which has |
| 3490 | // no concept of a single Isolate thread. Instead, it's OK for all inspector messages to be |
| 3491 | // dispatched on the calling thread. |
| 3492 | auto executor = kj::getCurrentThreadExecutor().addRef(); |
| 3493 | |
| 3494 | return attachInspector(kj::mv(executor), timer, timerOffset, *webSocket) |
| 3495 | .attach(kj::mv(webSocket)); |
| 3496 | } |
| 3497 | |
| 3498 | kj::Promise<void> Worker::Isolate::attachInspector( |
| 3499 | kj::Own<const kj::Executor> isolateThreadExecutor, |
| 3500 | kj::Timer& timer, |
| 3501 | kj::Duration timerOffset, |
| 3502 | kj::WebSocket& webSocket) const { |
| 3503 | KJ_REQUIRE(impl->inspector != kj::none); |
| 3504 | |
| 3505 | return jsg::runInV8Stack([&](jsg::V8StackScope& stackScope) { |
| 3506 | Isolate::Impl::Lock recordedLock( |
| 3507 | *this, InspectorChannelImpl::InspectorLock(kj::none), stackScope); |
| 3508 | auto& lock = *recordedLock.lock; |
| 3509 | auto& lockedSelf = const_cast<Worker::Isolate&>(*this); |
| 3510 | |
| 3511 | // If another inspector was already connected, boot it, on the assumption that that connection |
| 3512 | // is dead and this is why the user reconnected. While we could actually allow both inspector |
| 3513 | // sessions to stay open (V8 supports this!), we'd then need to store a set of all connected |
| 3514 | // inspectors in order to be able to disconnect all of them in case of an isolate purge... let's |
| 3515 | // just not. |
| 3516 | lockedSelf.disconnectInspector(); |
| 3517 | |
| 3518 | lockedSelf.impl->inspectorClient->setInspectorTimerInfo(timer, timerOffset); |
| 3519 | |
| 3520 | auto channel = kj::heap<Worker::Isolate::InspectorChannelImpl>( |
| 3521 | kj::atomicAddRef(*this), kj::mv(isolateThreadExecutor), webSocket); |
| 3522 | lockedSelf.currentInspectorSession = *channel; |
| 3523 | lockedSelf.impl->inspectorClient->setChannel(*channel); |
| 3524 | |
| 3525 | // Send any queued notifications. |
| 3526 | lock.withinHandleScope([&] { |
| 3527 | for (auto& notification: lockedSelf.impl->queuedNotifications) { |
| 3528 | channel->sendNotification(kj::mv(notification)); |
| 3529 | } |
| 3530 | lockedSelf.impl->queuedNotifications.clear(); |
| 3531 | }); |
| 3532 | |
| 3533 | return channel->messagePump().attach(kj::mv(channel)); |
| 3534 | }); |
| 3535 | } |
| 3536 | |
| 3537 | void Worker::Isolate::disconnectInspector() { |
| 3538 | // If an inspector session is connected, proactively drop it, so as to force it to drop its |
| 3539 | // reference on the script, so that the script can be deleted. |
| 3540 | KJ_IF_SOME(current, currentInspectorSession) { |
| 3541 | current.disconnect(); |
| 3542 | currentInspectorSession = kj::none; |
| 3543 | } |
| 3544 | impl->inspectorClient->resetChannel(); |
| 3545 | } |
| 3546 | |
| 3547 | void Worker::Isolate::logWarning(kj::StringPtr description, Lock& lock) { |
| 3548 | if (impl->inspector != kj::none) { |
| 3549 | JSG_WITHIN_CONTEXT_SCOPE(lock, lock.getContext(), [&](jsg::Lock& js) { |
| 3550 | logMessage(js, static_cast<uint16_t>(cdp::LogType::WARNING), description); |
| 3551 | }); |
| 3552 | } |
| 3553 | |
| 3554 | if (loggingOptions.consoleMode == Worker::ConsoleMode::INSPECTOR_ONLY) { |
| 3555 | // Run with --verbose to log JS exceptions to stderr. Useful when running tests. |
| 3556 | KJ_LOG(INFO, "console warning", description); |
| 3557 | } else { |
| 3558 | fprintf(stderr, "%s\n", description.cStr()); |
| 3559 | fflush(stderr); |
| 3560 | } |
| 3561 | |
| 3562 | KJ_IF_SOME(ioContext, IoContext::tryCurrent()) { |
| 3563 | KJ_IF_SOME(tracer, ioContext.getWorkerTracer()) { |
| 3564 | // json encoding is required over simply wrapping it in quotes to correctly escape the string. |
| 3565 | capnp::JsonCodec json; |
| 3566 | auto jsonDescription = kj::str("[", json.encode(capnp::Text::Reader(description)), "]"); |
| 3567 | |
| 3568 | auto timestamp = ioContext.now(); |
| 3569 | tracer.addLog( |
| 3570 | ioContext.getInvocationSpanContext(), timestamp, LogLevel::WARN, kj::mv(jsonDescription)); |
| 3571 | } |
| 3572 | } |
| 3573 | } |
| 3574 | |
| 3575 | void Worker::Isolate::logWarningOnce(kj::StringPtr description, Lock& lock) { |
| 3576 | impl->warningOnceDescriptions.findOrCreate(description, [&] { |
| 3577 | logWarning(description, lock); |
| 3578 | return kj::str(description); |
| 3579 | }); |
| 3580 | } |
| 3581 | |
| 3582 | void Worker::Isolate::logErrorOnce(kj::StringPtr description) { |
| 3583 | impl->errorOnceDescriptions.findOrCreate(description, [&] { |
| 3584 | KJ_LOG(ERROR, description); |
| 3585 | return kj::str(description); |
| 3586 | }); |
| 3587 | } |
| 3588 | |
| 3589 | void Worker::Isolate::logMessage(jsg::Lock& js, uint16_t type, kj::StringPtr description) { |
| 3590 | if (impl->inspector != kj::none) { |
| 3591 | // We want to log a warning to the devtools console, as if `console.warn()` were called. |
| 3592 | // However, the only public interface to call the real `console.warn()` is via JavaScript, |
| 3593 | // where it could have been monkey-patched by the guest. We'd like to avoid having to worry |
| 3594 | // about that blowing up in our face. So instead we arrange to send the proper devtools |
| 3595 | // protocol messages ourselves. |
| 3596 | // |
| 3597 | // TODO(cleanup): It would be better if we could directly add the message to the inspector's |
| 3598 | // console log (without calling through JavaScript). What we're doing here has some problems. |
| 3599 | // In particular, if no client is connected yet, we attempt to queue up the messages to send |
| 3600 | // later, much like the real inspector does. This is kind of complicated, and doesn't quite |
| 3601 | // work right: |
| 3602 | // - The messages won't necessarily be in the right order with normal console logs made at |
| 3603 | // the same time (with identical timestamps). |
| 3604 | // - In theory we should queue *all* logged warnings and deliver them to every future client, |
| 3605 | // not just the next client to connect. But if we do that, we also need to respect the |
| 3606 | // protocol command to clear the history when requested. This was further than I cared to |
| 3607 | // go. |
| 3608 | // To fix these problems, maybe we should just patch V8 with a direct interface into the |
| 3609 | // inspector's own log. (Also, how does Chrome handle this?) |
| 3610 | |
| 3611 | js.withinHandleScope([&] { |
| 3612 | capnp::MallocMessageBuilder message; |
| 3613 | auto event = message.initRoot<cdp::Event>(); |
| 3614 | |
| 3615 | auto params = event.initRuntimeConsoleApiCalled(); |
| 3616 | params.setType(static_cast<cdp::LogType>(type)); |
| 3617 | params.initArgs(1)[0].initString().setValue(description); |
| 3618 | params.setExecutionContextId(v8_inspector::V8ContextInfo::executionContextId(js.v8Context())); |
| 3619 | params.setTimestamp(impl->inspectorClient->currentTimeMS()); |
| 3620 | stackTraceToCDP(js, params.initStackTrace()); |
| 3621 | |
| 3622 | auto notification = getCdpJsonCodec().encode(event); |
| 3623 | KJ_IF_SOME(i, currentInspectorSession) { |
| 3624 | i.sendNotification(kj::mv(notification)); |
| 3625 | } else { |
| 3626 | impl->queuedNotifications.add(kj::mv(notification)); |
| 3627 | } |
| 3628 | }); |
| 3629 | } |
| 3630 | } |
| 3631 | |
| 3632 | // ======================================================================================= |
| 3633 | |
| 3634 | struct Worker::Actor::Impl { |
| 3635 | Actor::Id actorId; |
| 3636 | Frankenvalue props; |
| 3637 | MakeStorageFunc makeStorage; |
| 3638 | |
| 3639 | kj::Own<ActorObserver> metrics; |
| 3640 | |
| 3641 | // When a boolean, indicates whether a `transient` should exist. If true, it will be initialized |
| 3642 | // on the first `ensureConstructed()`. |
| 3643 | kj::OneOf<bool, jsg::JsRef<jsg::JsValue>> transient; |
| 3644 | |
| 3645 | kj::Maybe<kj::Own<ActorCacheInterface>> actorCache; |
| 3646 | |
| 3647 | kj::Maybe<jsg::JsRef<jsg::JsObject>> ctxObject; |
| 3648 | |
| 3649 | kj::Maybe<rpc::Container::Client> container; |
| 3650 | kj::Maybe<FacetManager&> facetManager; |
| 3651 | kj::Maybe<ActorVersion> version; |
| 3652 | |
| 3653 | struct NoClass {}; |
| 3654 | struct Initializing {}; |
| 3655 | |
| 3656 | // If the actor is backed by a class, this field tracks the instance through its stages. The |
| 3657 | // instance is constructed as part of the first request to be delivered. |
| 3658 | kj::OneOf<NoClass, // not class-based |
| 3659 | Worker::ActorClassInfo*, // constructor not run yet |
| 3660 | Initializing, // constructor currently running |
| 3661 | api::ExportedHandler, // fully constructed |
| 3662 | kj::Exception // constructor threw |
| 3663 | > |
| 3664 | classInstance; |
| 3665 | |
| 3666 | class HooksImpl: public InputGate::Hooks, public OutputGate::Hooks, public ActorCache::Hooks { |
| 3667 | public: |
| 3668 | HooksImpl(kj::Own<Loopback> loopback, TimerChannel& timerChannel, ActorObserver& metrics) |
| 3669 | : loopback(kj::mv(loopback)), |
| 3670 | timerChannel(timerChannel), |
| 3671 | metrics(metrics) {} |
| 3672 | |
| 3673 | void inputGateLocked() override { |
| 3674 | metrics.inputGateLocked(); |
| 3675 | } |
| 3676 | void inputGateReleased() override { |
| 3677 | metrics.inputGateReleased(); |
| 3678 | } |
| 3679 | void inputGateWaiterAdded() override { |
| 3680 | metrics.inputGateWaiterAdded(); |
| 3681 | } |
| 3682 | void inputGateWaiterRemoved() override { |
| 3683 | metrics.inputGateWaiterRemoved(); |
| 3684 | } |
| 3685 | // Implements InputGate::Hooks. |
| 3686 | |
| 3687 | kj::Promise<void> makeTimeoutPromise() override { |
| 3688 | // This really only protects against total hangs. Lowering the timeout drastically is risky, |
| 3689 | // since low timeouts can spuriously fire when under heavy CPU load, failing requests that |
| 3690 | // would otherwise succeed. |
| 3691 | auto timeout = 30 * kj::SECONDS; |
| 3692 | co_await timerChannel.afterLimitTimeout(timeout); |
| 3693 | |
| 3694 | kj::throwFatalException(KJ_EXCEPTION(OVERLOADED, |
| 3695 | "broken.outputGateBroken; jsg.Error: Durable Object storage operation exceeded " |
| 3696 | "timeout which caused object to be reset.")); |
| 3697 | } |
| 3698 | |
| 3699 | // Implements OutputGate::Hooks. |
| 3700 | |
| 3701 | void outputGateLocked() override { |
| 3702 | metrics.outputGateLocked(); |
| 3703 | } |
| 3704 | void outputGateReleased() override { |
| 3705 | metrics.outputGateReleased(); |
| 3706 | } |
| 3707 | void outputGateWaiterAdded() override { |
| 3708 | metrics.outputGateWaiterAdded(); |
| 3709 | } |
| 3710 | void outputGateWaiterRemoved() override { |
| 3711 | metrics.outputGateWaiterRemoved(); |
| 3712 | } |
| 3713 | |
| 3714 | // Implements ActorCache::Hooks |
| 3715 | |
| 3716 | void updateAlarmInMemory(kj::Maybe<kj::Date> newAlarmTime) override; |
| 3717 | void storageReadCompleted(kj::Duration latency) override { |
| 3718 | metrics.storageReadCompleted(latency); |
| 3719 | } |
| 3720 | void storageWriteCompleted(kj::Duration latency) override { |
| 3721 | metrics.storageWriteCompleted(latency); |
| 3722 | } |
| 3723 | |
| 3724 | private: |
| 3725 | kj::Own<Loopback> loopback; // only for updateAlarmInMemory() |
| 3726 | TimerChannel& timerChannel; // only for afterLimitTimeout() and updateAlarmInMemory() |
| 3727 | ActorObserver& metrics; |
| 3728 | |
| 3729 | kj::Maybe<kj::Promise<void>> maybeAlarmPreviewTask; |
| 3730 | }; |
| 3731 | |
| 3732 | HooksImpl hooks; |
| 3733 | |
| 3734 | // Handles both input locks and request locks. |
| 3735 | InputGate inputGate; |
| 3736 | |
| 3737 | // Handles output locks. |
| 3738 | OutputGate outputGate; |
| 3739 | |
| 3740 | // `ioContext` is initialized upon delivery of the first request. |
| 3741 | kj::Maybe<kj::Own<IoContext>> ioContext; |
| 3742 | |
| 3743 | // If onBroken() is called while `ioContext` is still null, this is initialized. When |
| 3744 | // `ioContext` is constructed, this will be fulfilled with `ioContext.onAbort()`. |
| 3745 | kj::Maybe<kj::Own<kj::PromiseFulfiller<kj::Promise<void>>>> abortFulfiller; |
| 3746 | |
| 3747 | // Task which periodically flushes metrics. Initialized after `ioContext` is initialized. |
| 3748 | kj::Maybe<kj::Promise<void>> metricsFlushLoopTask; |
| 3749 | |
| 3750 | // Allows sending requests back into this actor, recreating it as necessary. Safe to hold longer |
| 3751 | // than the Worker::Actor is alive. |
| 3752 | kj::Own<Loopback> loopback; |
| 3753 | |
| 3754 | TimerChannel& timerChannel; |
| 3755 | |
| 3756 | kj::ForkedPromise<void> shutdownPromise; |
| 3757 | kj::Own<kj::PromiseFulfiller<void>> shutdownFulfiller; |
| 3758 | |
| 3759 | // If this Actor has a HibernationManager, it means the Actor has recently accepted a Hibernatable |
| 3760 | // websocket. We eventually move the HibernationManager into the DeferredProxy task |
| 3761 | // (since it's long lived), but can still refer to the HibernationManager by passing a reference |
| 3762 | // in each CustomEvent. |
| 3763 | kj::Maybe<kj::Own<HibernationManager>> hibernationManager; |
| 3764 | kj::Maybe<uint16_t> hibernationEventType; |
| 3765 | |
| 3766 | struct ScheduledAlarm { |
| 3767 | ScheduledAlarm( |
| 3768 | kj::Date scheduledTime, kj::PromiseFulfillerPair<WorkerInterface::AlarmOutcome> pf) |
| 3769 | : scheduledTime(scheduledTime), |
| 3770 | resultFulfiller(kj::mv(pf.fulfiller)), |
| 3771 | resultPromise(pf.promise.fork()) {} |
| 3772 | KJ_DISALLOW_COPY(ScheduledAlarm); |
| 3773 | ScheduledAlarm(ScheduledAlarm&&) = default; |
| 3774 | ~ScheduledAlarm() noexcept(false) {} |
| 3775 | |
| 3776 | kj::Date scheduledTime; |
| 3777 | WorkerInterface::AlarmFulfiller resultFulfiller; |
| 3778 | kj::ForkedPromise<WorkerInterface::AlarmOutcome> resultPromise; |
| 3779 | kj::Promise<void> cleanupPromise = resultPromise.addBranch().then( |
| 3780 | [](WorkerInterface::AlarmOutcome&&) {}, [](kj::Exception&&) {}); |
| 3781 | // The first thing we do after we get a result should be to remove the running alarm (if we got |
| 3782 | // that far). So we grab the first branch now and ignore any results, before anyone else has a |
| 3783 | // chance to do so. |
| 3784 | }; |
| 3785 | struct RunningAlarm { |
| 3786 | kj::Date scheduledTime; |
| 3787 | kj::ForkedPromise<WorkerInterface::AlarmOutcome> resultPromise; |
| 3788 | }; |
| 3789 | // If valid, we have an alarm invocation that has not yet received an `AlarmFulfiller` and thus |
| 3790 | // is either waiting for a running alarm or its scheduled time. |
| 3791 | kj::Maybe<ScheduledAlarm> maybeScheduledAlarm; |
| 3792 | |
| 3793 | // If valid, we have an alarm invocation that has received an `AlarmFulfiller` and is currently |
| 3794 | // considered running. This alarm is no longer cancelable. |
| 3795 | kj::Maybe<RunningAlarm> maybeRunningAlarm; |
| 3796 | |
| 3797 | // This is a forked promise so that we can schedule and then cancel multiple alarms while an alarm |
| 3798 | // is running. |
| 3799 | kj::ForkedPromise<void> runningAlarmTask = kj::Promise<void>(kj::READY_NOW).fork(); |
| 3800 | |
| 3801 | Impl(Worker::Actor& self, |
| 3802 | Actor::Id actorId, |
| 3803 | bool hasTransient, |
| 3804 | MakeActorCacheFunc makeActorCache, |
| 3805 | Frankenvalue props, |
| 3806 | MakeStorageFunc makeStorage, |
| 3807 | kj::Own<Loopback> loopback, |
| 3808 | TimerChannel& timerChannel, |
| 3809 | kj::Own<ActorObserver> metricsParam, |
| 3810 | kj::Maybe<kj::Own<HibernationManager>> manager, |
| 3811 | kj::Maybe<uint16_t>& hibernationEventType, |
| 3812 | kj::Maybe<rpc::Container::Client> container, |
| 3813 | kj::Maybe<FacetManager&> facetManager, |
| 3814 | kj::PromiseFulfillerPair<void> paf = kj::newPromiseAndFulfiller<void>()) |
| 3815 | : actorId(kj::mv(actorId)), |
| 3816 | props(kj::mv(props)), |
| 3817 | makeStorage(kj::mv(makeStorage)), |
| 3818 | metrics(kj::mv(metricsParam)), |
| 3819 | transient(hasTransient), |
| 3820 | container(kj::mv(container)), |
| 3821 | facetManager(facetManager), |
| 3822 | hooks(loopback->addRef(), timerChannel, *metrics), |
| 3823 | inputGate(hooks), |
| 3824 | outputGate(hooks), |
| 3825 | loopback(kj::mv(loopback)), |
| 3826 | timerChannel(timerChannel), |
| 3827 | shutdownPromise(paf.promise.fork()), |
| 3828 | shutdownFulfiller(kj::mv(paf.fulfiller)), |
| 3829 | hibernationManager(kj::mv(manager)), |
| 3830 | hibernationEventType(kj::mv(hibernationEventType)) { |
| 3831 | actorCache = |
| 3832 | makeActorCache(self.worker->getIsolate().impl->actorCacheLru, outputGate, hooks, *metrics); |
| 3833 | } |
| 3834 | }; |
| 3835 | |
| 3836 | kj::Promise<Worker::AsyncLock> Worker::takeAsyncLockWhenActorCacheReady( |
| 3837 | kj::Date now, Actor& actor, RequestObserver& request) const { |
| 3838 | auto lockTiming = |
| 3839 | getIsolate().getMetrics().tryCreateLockTiming(kj::Maybe<RequestObserver&>(request)); |
| 3840 | |
| 3841 | KJ_IF_SOME(c, actor.impl->actorCache) { |
| 3842 | KJ_IF_SOME(p, c.get()->evictStale(now)) { |
| 3843 | // Got backpressure, wait for it. |
| 3844 | // TODO(someday): Count this time period differently in lock timing data? |
| 3845 | co_await p; |
| 3846 | } |
| 3847 | } |
| 3848 | |
| 3849 | co_return co_await getIsolate().takeAsyncLockImpl(kj::mv(lockTiming)); |
| 3850 | } |
| 3851 | |
| 3852 | Worker::Actor::Actor(const Worker& worker, |
| 3853 | kj::Maybe<RequestTracker&> tracker, |
| 3854 | Actor::Id actorId, |
| 3855 | bool hasTransient, |
| 3856 | MakeActorCacheFunc makeActorCache, |
| 3857 | kj::Maybe<kj::StringPtr> className, |
| 3858 | Frankenvalue props, |
| 3859 | MakeStorageFunc makeStorage, |
| 3860 | kj::Own<Loopback> loopback, |
| 3861 | TimerChannel& timerChannel, |
| 3862 | kj::Own<ActorObserver> metrics, |
| 3863 | kj::Maybe<kj::Own<HibernationManager>> manager, |
| 3864 | kj::Maybe<uint16_t> hibernationEventType, |
| 3865 | kj::Maybe<rpc::Container::Client> container, |
| 3866 | kj::Maybe<FacetManager&> facetManager, |
| 3867 | kj::Maybe<ActorVersion> version) |
| 3868 | : worker(kj::atomicAddRef(worker)), |
| 3869 | tracker(tracker.map([](RequestTracker& tracker) { return tracker.addRef(); })) { |
| 3870 | impl = kj::heap<Impl>(*this, kj::mv(actorId), hasTransient, kj::mv(makeActorCache), kj::mv(props), |
| 3871 | kj::mv(makeStorage), kj::mv(loopback), timerChannel, kj::mv(metrics), kj::mv(manager), |
| 3872 | hibernationEventType, kj::mv(container), facetManager); |
| 3873 | impl->version = kj::mv(version); |
| 3874 | |
| 3875 | KJ_IF_SOME(c, className) { |
| 3876 | KJ_IF_SOME(cls, worker.impl->actorClasses.find(c)) { |
| 3877 | // const_cast OK because we're just storing the pointer and will only use this under lock. |
| 3878 | impl->classInstance = const_cast<ActorClassInfo*>(&cls); |
| 3879 | } else { |
| 3880 | auto e = KJ_EXCEPTION(FAILED, "broken.ignored; no such actor class", c); |
| 3881 | e.setDetail(jsg::EXCEPTION_IS_USER_ERROR, kj::heapArray<kj::byte>(0)); |
| 3882 | kj::throwFatalException(kj::mv(e)); |
| 3883 | } |
| 3884 | } else { |
| 3885 | impl->classInstance = Impl::NoClass(); |
| 3886 | } |
| 3887 | } |
| 3888 | |
| 3889 | void Worker::Actor::ensureConstructed(IoContext& context) { |
| 3890 | KJ_IF_SOME(info, impl->classInstance.tryGet<ActorClassInfo*>()) { |
| 3891 | // IMPORTANT: We need to set the state to "Initializing" synchronously, before |
| 3892 | // ensureConstructedImpl() actually executes and acquires the input lock. |
| 3893 | // This prevents multiple concurrent initialization attempts if multiple calls to |
| 3894 | // ensureConstructed() arrive back-to-back. |
| 3895 | // |
| 3896 | // This doesn't create a race condition with getHandler() because InputGate::wait() |
| 3897 | // synchronously adds the caller to the wait queue, even though it completes |
| 3898 | // asynchronously. Any call to getHandler() that arrives after this point will |
| 3899 | // have to wait for the input lock, which is only acquired and released by |
| 3900 | // ensureConstructedImpl() when it completes initialization. |
| 3901 | // |
| 3902 | // So the "actor still initializing" error in getHandler() should be impossible |
| 3903 | // unless a code path is bypassing the input lock mechanism. |
| 3904 | context.addWaitUntil(ensureConstructedImpl(context, *info)); |
| 3905 | impl->classInstance = Impl::Initializing(); |
| 3906 | } |
| 3907 | } |
| 3908 | |
| 3909 | kj::Promise<void> Worker::Actor::ensureConstructedImpl(IoContext& context, ActorClassInfo& info) { |
| 3910 | InputGate::Lock inputLock = co_await impl->inputGate.wait(context.getCurrentTraceSpan()); |
| 3911 | |
| 3912 | try { |
| 3913 | bool containerRunning = false; |
| 3914 | KJ_IF_SOME(c, impl->container) { |
| 3915 | // We need to do an RPC to check if the container is running. |
| 3916 | // TODO(perf): It would be nice if we could have started this RPC earlier, e.g. in parallel |
| 3917 | // with starting the script, and also if we could save the status across hibernations. But |
| 3918 | // that would require some refactoring, and this RPC should (eventally) be local, so it's |
| 3919 | // not a huge deal. |
| 3920 | auto status = co_await c.statusRequest(capnp::MessageSize{4, 0}).send(); |
| 3921 | containerRunning = status.getRunning(); |
| 3922 | } |
| 3923 | |
| 3924 | co_await context.run([this, &info, containerRunning](Worker::Lock& lock) { |
| 3925 | jsg::Lock& js = lock; |
| 3926 | |
| 3927 | kj::Maybe<jsg::Ref<api::DurableObjectStorage>> storage; |
| 3928 | KJ_IF_SOME(c, impl->actorCache) { |
| 3929 | storage = impl->makeStorage(lock, worker->getIsolate().getApi(), *c); |
| 3930 | } |
| 3931 | |
| 3932 | auto ctx = js.alloc<api::DurableObjectState>(js, cloneId(), |
| 3933 | jsg::JsValue(KJ_ASSERT_NONNULL(lock.getWorker().impl->ctxExports).getHandle(js)), |
| 3934 | impl->props.toJs(js), kj::mv(storage), kj::mv(impl->container), containerRunning, |
| 3935 | impl->facetManager, impl->version.map([](ActorVersion& v) { |
| 3936 | return ActorVersion{.cohort = v.cohort.map([](kj::String& s) { return kj::str(s); })}; |
| 3937 | })); |
| 3938 | |
| 3939 | auto handler = |
| 3940 | info.cls(lock, ctx.addRef(), KJ_ASSERT_NONNULL(lock.getWorker().impl->env).addRef(js)); |
| 3941 | |
| 3942 | // Since we JUST passed `ctx` into the class constructor, it definitely has a handle |
| 3943 | // attached. Let's grab it and stash it to implement getCtx(). |
| 3944 | auto ctxHandle = jsg::JsObject(KJ_ASSERT_NONNULL(ctx.tryGetHandle(js))); |
| 3945 | impl->ctxObject = jsg::JsRef<jsg::JsObject>(js, ctxHandle); |
| 3946 | |
| 3947 | // HACK: We set handler.env to undefined because we already passed the real env into the |
| 3948 | // constructor, and we want the handler methods to act like they take just one parameter. |
| 3949 | // We do the same for handler.ctx, as ExecutionContext related tasks are performed |
| 3950 | // on the actor's state field instead. |
| 3951 | handler.env = js.v8Ref(js.v8Undefined()); |
| 3952 | handler.ctx = kj::none; |
| 3953 | handler.missingSuperclass = info.missingSuperclass; |
| 3954 | |
| 3955 | impl->classInstance = kj::mv(handler); |
| 3956 | }, inputLock.addRef(context.getCurrentTraceSpan())); |
| 3957 | // We addRef() the inputLock above rather than kj::mv() it so that the lock remains held |
| 3958 | // through the catch block below, if an exception is thrown. This is important since we |
| 3959 | // MUST update `impl->classInstance` to something other than `Initializing` before we |
| 3960 | // release the lock. |
| 3961 | } catch (...) { |
| 3962 | // Get the KJ exception |
| 3963 | auto e = kj::getCaughtExceptionAsKj(); |
| 3964 | |
| 3965 | auto msg = e.getDescription(); |
| 3966 | if (!msg.startsWith("broken."_kj) && !msg.startsWith("remote.broken."_kj)) { |
| 3967 | // If we already set up a brokenness reason, we shouldn't override it. |
| 3968 | auto description = jsg::annotateBroken(msg, "broken.constructorFailed"); |
| 3969 | e.setDescription(kj::mv(description)); |
| 3970 | } |
| 3971 | |
| 3972 | context.abort(e.clone()); |
| 3973 | impl->classInstance = kj::mv(e); |
| 3974 | } |
| 3975 | } |
| 3976 | |
| 3977 | Worker::Actor::~Actor() noexcept(false) { |
| 3978 | // Note: We do not need an isolate lock to destroy the actor impl. Everything in it is specific |
| 3979 | // to our thread, or is a handle that can be dropped outside of the lock. |
| 3980 | } |
| 3981 | |
| 3982 | void Worker::Actor::shutdown(uint16_t reasonCode, kj::Maybe<const kj::Exception&> error) { |
| 3983 | // We're officially canceling all background work and we're going to destruct the Actor as soon |
| 3984 | // as all IoContexts that reference it go out of scope. We might still log additional |
| 3985 | // periodic messages, and that's good because we might care about that information. That said, |
| 3986 | // we're officially "broken" from this point because we cannot service background work and our |
| 3987 | // capability server should have triggered this (potentially indirectly) via its destructor. |
| 3988 | KJ_IF_SOME(r, impl->ioContext) { |
| 3989 | impl->metrics->shutdown(reasonCode, r.get()->getLimitEnforcer()); |
| 3990 | } else { |
| 3991 | // The actor was shut down before the IoContext was even constructed, so no metrics are |
| 3992 | // written. |
| 3993 | } |
| 3994 | |
| 3995 | shutdownActorCache(error); |
| 3996 | |
| 3997 | impl->shutdownFulfiller->fulfill(); |
| 3998 | } |
| 3999 | |
| 4000 | void Worker::Actor::shutdownActorCache(kj::Maybe<const kj::Exception&> error) { |
| 4001 | KJ_IF_SOME(ac, impl->actorCache) { |
| 4002 | ac.get()->shutdown(error); |
| 4003 | } else { |
| 4004 | // The actor was aborted before the actor cache was constructed, nothing to do. |
| 4005 | } |
| 4006 | } |
| 4007 | |
| 4008 | kj::Promise<void> Worker::Actor::onShutdown() { |
| 4009 | return impl->shutdownPromise.addBranch(); |
| 4010 | } |
| 4011 | |
| 4012 | kj::Promise<void> Worker::Actor::onBroken() { |
| 4013 | // TODO(soon): Detect and report other cases of brokenness, as described in worker.capnp. |
| 4014 | |
| 4015 | kj::Promise<void> abortPromise = nullptr; |
| 4016 | |
| 4017 | KJ_IF_SOME(rc, impl->ioContext) { |
| 4018 | abortPromise = rc.get()->onAbort(); |
| 4019 | } else { |
| 4020 | auto paf = kj::newPromiseAndFulfiller<kj::Promise<void>>(); |
| 4021 | abortPromise = kj::mv(paf.promise); |
| 4022 | impl->abortFulfiller = kj::mv(paf.fulfiller); |
| 4023 | } |
| 4024 | |
| 4025 | return abortPromise; |
| 4026 | } |
| 4027 | |
| 4028 | const Worker::Actor::Id& Worker::Actor::getId() { |
| 4029 | return impl->actorId; |
| 4030 | } |
| 4031 | |
| 4032 | bool Worker::Actor::idsEqual(const Id& a, const Id& b) { |
| 4033 | if (a.which() != b.which()) return false; |
| 4034 | |
| 4035 | KJ_SWITCH_ONEOF(a) { |
| 4036 | KJ_CASE_ONEOF(actorId, kj::Own<ActorIdFactory::ActorId>) { |
| 4037 | return actorId->equals(*b.get<kj::Own<ActorIdFactory::ActorId>>()); |
| 4038 | } |
| 4039 | KJ_CASE_ONEOF(str, kj::String) { |
| 4040 | return str == b.get<kj::String>(); |
| 4041 | } |
| 4042 | } |
| 4043 | KJ_UNREACHABLE; |
| 4044 | } |
| 4045 | |
| 4046 | Worker::Actor::Id Worker::Actor::cloneId(Worker::Actor::Id& id) { |
| 4047 | KJ_SWITCH_ONEOF(id) { |
| 4048 | KJ_CASE_ONEOF(coloLocalId, kj::String) { |
| 4049 | return kj::str(coloLocalId); |
| 4050 | } |
| 4051 | KJ_CASE_ONEOF(globalId, kj::Own<ActorIdFactory::ActorId>) { |
| 4052 | return globalId->clone(); |
| 4053 | } |
| 4054 | } |
| 4055 | KJ_UNREACHABLE; |
| 4056 | } |
| 4057 | |
| 4058 | Worker::Actor::Id Worker::Actor::cloneId() { |
| 4059 | return cloneId(impl->actorId); |
| 4060 | } |
| 4061 | |
| 4062 | kj::Maybe<jsg::JsRef<jsg::JsValue>> Worker::Actor::getTransient(Worker::Lock& lock) { |
| 4063 | KJ_REQUIRE(&lock.getWorker() == worker.get()); |
| 4064 | |
| 4065 | if (impl->transient.tryGet<bool>().orDefault(false)) { |
| 4066 | // First call and `hasTransient` was true. Initialize it now, since we have the lock. |
| 4067 | jsg::Lock& js = lock; |
| 4068 | impl->transient.init<jsg::JsRef<jsg::JsValue>>(js, js.obj()); |
| 4069 | } |
| 4070 | |
| 4071 | return impl->transient.tryGet<jsg::JsRef<jsg::JsValue>>().map( |
| 4072 | [&](jsg::JsRef<jsg::JsValue>& val) { return val.addRef(lock); }); |
| 4073 | } |
| 4074 | |
| 4075 | kj::Maybe<ActorCacheInterface&> Worker::Actor::getPersistent() { |
| 4076 | return impl->actorCache; |
| 4077 | } |
| 4078 | |
| 4079 | kj::Own<Worker::Actor::Loopback> Worker::Actor::getLoopback() { |
| 4080 | return impl->loopback->addRef(); |
| 4081 | } |
| 4082 | |
| 4083 | kj::Maybe<jsg::Ref<api::DurableObjectStorage>> Worker::Actor::makeStorageForSwSyntax( |
| 4084 | Worker::Lock& lock) { |
| 4085 | return impl->actorCache.map([&](kj::Own<ActorCacheInterface>& cache) { |
| 4086 | return impl->makeStorage(lock, worker->getIsolate().getApi(), *cache); |
| 4087 | }); |
| 4088 | } |
| 4089 | |
| 4090 | void Worker::Actor::assertCanSetAlarm() { |
| 4091 | KJ_SWITCH_ONEOF(impl->classInstance) { |
| 4092 | KJ_CASE_ONEOF(_, Impl::NoClass) { |
| 4093 | // Once upon a time, we allowed actors without classes. Let's make a nicer message if we |
| 4094 | // we somehow see a classless actor attempt to run an alarm in the wild. |
| 4095 | JSG_FAIL_REQUIRE( |
| 4096 | TypeError, "Your Durable Object must be class-based in order to call setAlarm()"); |
| 4097 | } |
| 4098 | KJ_CASE_ONEOF(_, Worker::ActorClassInfo*) { |
| 4099 | KJ_FAIL_ASSERT("setAlarm() invoked before Durable Object ctor"); |
| 4100 | } |
| 4101 | KJ_CASE_ONEOF(_, Impl::Initializing) { |
| 4102 | // We don't explicitly know if we have an alarm handler or not, so just let it happen. We'll |
| 4103 | // handle it when we go to run the alarm. |
| 4104 | return; |
| 4105 | } |
| 4106 | KJ_CASE_ONEOF(handler, api::ExportedHandler) { |
| 4107 | JSG_REQUIRE(handler.alarm != kj::none, TypeError, |
| 4108 | "Your Durable Object class must have an alarm() handler in order to call setAlarm()"); |
| 4109 | return; |
| 4110 | } |
| 4111 | KJ_CASE_ONEOF(exception, kj::Exception) { |
| 4112 | // We've failed in the ctor, might as well just throw that exception for now. |
| 4113 | kj::throwFatalException(exception.clone()); |
| 4114 | } |
| 4115 | } |
| 4116 | KJ_UNREACHABLE; |
| 4117 | } |
| 4118 | |
| 4119 | void Worker::Actor::Impl::HooksImpl::updateAlarmInMemory(kj::Maybe<kj::Date> newTime) { |
| 4120 | if (newTime == kj::none) { |
| 4121 | maybeAlarmPreviewTask = kj::none; |
| 4122 | return; |
| 4123 | } |
| 4124 | |
| 4125 | auto scheduledTime = KJ_ASSERT_NONNULL(newTime); |
| 4126 | |
| 4127 | auto retry = kj::coCapture([this, originalTime = scheduledTime]() -> kj::Promise<void> { |
| 4128 | kj::Date scheduledTime = originalTime; |
| 4129 | |
| 4130 | for (auto i: kj::zeroTo(WorkerInterface::ALARM_RETRY_MAX_TRIES)) { |
| 4131 | co_await timerChannel.atTime(scheduledTime); |
| 4132 | auto result = co_await loopback->getWorker(IoChannelFactory::SubrequestMetadata{}) |
| 4133 | ->runAlarm(originalTime, i); |
| 4134 | |
| 4135 | if (result.outcome == EventOutcome::OK || !result.retry) { |
| 4136 | break; |
| 4137 | } |
| 4138 | |
| 4139 | auto delay = (WorkerInterface::ALARM_RETRY_START_SECONDS << i++) * kj::SECONDS; |
| 4140 | scheduledTime = timerChannel.now() + delay; |
| 4141 | } |
| 4142 | }); |
| 4143 | |
| 4144 | maybeAlarmPreviewTask = retry(); |
| 4145 | } |
| 4146 | |
| 4147 | kj::Maybe<kj::Promise<WorkerInterface::AlarmOutcome>> Worker::Actor::getAlarm( |
| 4148 | kj::Date scheduledTime) { |
| 4149 | KJ_IF_SOME(runningAlarm, impl->maybeRunningAlarm) { |
| 4150 | if (runningAlarm.scheduledTime == scheduledTime) { |
| 4151 | // The running alarm has the same time, we can just wait for it. |
| 4152 | return runningAlarm.resultPromise.addBranch(); |
| 4153 | } |
| 4154 | } |
| 4155 | |
| 4156 | KJ_IF_SOME(scheduledAlarm, impl->maybeScheduledAlarm) { |
| 4157 | if (scheduledAlarm.scheduledTime == scheduledTime) { |
| 4158 | // The scheduled alarm has the same time, we can just wait for it. |
| 4159 | return scheduledAlarm.resultPromise.addBranch(); |
| 4160 | } |
| 4161 | } |
| 4162 | |
| 4163 | return kj::none; |
| 4164 | } |
| 4165 | |
| 4166 | kj::Promise<WorkerInterface::ScheduleAlarmResult> Worker::Actor::scheduleAlarm( |
| 4167 | kj::Date scheduledTime) { |
| 4168 | KJ_IF_SOME(runningAlarm, impl->maybeRunningAlarm) { |
| 4169 | if (runningAlarm.scheduledTime == scheduledTime) { |
| 4170 | // The running alarm has the same time, we can just wait for it. |
| 4171 | auto result = co_await runningAlarm.resultPromise; |
| 4172 | co_return result; |
| 4173 | } |
| 4174 | } |
| 4175 | |
| 4176 | KJ_IF_SOME(scheduledAlarm, impl->maybeScheduledAlarm) { |
| 4177 | // We had a previously scheduled alarm, let's cancel it. |
| 4178 | scheduledAlarm.resultFulfiller.cancel(); |
| 4179 | impl->maybeScheduledAlarm = kj::none; |
| 4180 | } |
| 4181 | |
| 4182 | KJ_IASSERT(impl->maybeScheduledAlarm == kj::none); |
| 4183 | auto& scheduledAlarm = impl->maybeScheduledAlarm.emplace( |
| 4184 | scheduledTime, kj::newPromiseAndFulfiller<WorkerInterface::AlarmOutcome>()); |
| 4185 | |
| 4186 | // Probably don't need to use kj::coCapture for this but doing so just to be on the |
| 4187 | // safe side... |
| 4188 | auto whenCanceled = |
| 4189 | (kj::coCapture([&scheduledAlarm]() -> kj::Promise<WorkerInterface::ScheduleAlarmResult> { |
| 4190 | // We've been cancelled, so return that result. Note that we cannot be resolved any other |
| 4191 | // way until we return an AlarmFulfiller below. |
| 4192 | co_return co_await scheduledAlarm.resultPromise; |
| 4193 | }))(); |
| 4194 | |
| 4195 | // Date.now() < scheduledTime when the alarm comes in, since we subtract elapsed CPU time from |
| 4196 | // the time of last I/O in the implementation of Date.now(). This difference could be used to |
| 4197 | // implement a Spectre timer, so we have to wait a little longer until |
| 4198 | // `Date.now() == scheduledTime`. Note that this also means that we could invoke ahead of its |
| 4199 | // `scheduledTime` and we'll delay until appropriate, this may be useful in cases of clock skew. |
| 4200 | |
| 4201 | co_return co_await handleAlarm(scheduledTime).exclusiveJoin(kj::mv(whenCanceled)); |
| 4202 | } |
| 4203 | |
| 4204 | kj::Promise<WorkerInterface::ScheduleAlarmResult> Worker::Actor::handleAlarm( |
| 4205 | kj::Date scheduledTime) { |
| 4206 | // Let's wait for any running alarm to cleanup before we even delay. |
| 4207 | co_await impl->runningAlarmTask; |
| 4208 | |
| 4209 | co_await KJ_ASSERT_NONNULL(impl->ioContext)->atTime(scheduledTime); |
| 4210 | // It's time to run! Let's tear apart the scheduled alarm and make a running alarm. |
| 4211 | |
| 4212 | // `maybeScheduledAlarm` should have the same value we emplaced above. If another call to |
| 4213 | // `scheduleAlarm()` emplaced a new value, then `whenCanceled` should have resolved which |
| 4214 | // cancels this this promise chain. |
| 4215 | auto scheduledAlarm = KJ_ASSERT_NONNULL(kj::mv(impl->maybeScheduledAlarm)); |
| 4216 | impl->maybeScheduledAlarm = kj::none; |
| 4217 | |
| 4218 | impl->maybeRunningAlarm.emplace(Impl::RunningAlarm{ |
| 4219 | .scheduledTime = scheduledAlarm.scheduledTime, |
| 4220 | .resultPromise = kj::mv(scheduledAlarm.resultPromise), |
| 4221 | }); |
| 4222 | impl->runningAlarmTask = scheduledAlarm.cleanupPromise |
| 4223 | .attach(kj::defer([&impl = *impl]() { |
| 4224 | // As soon as we get fulfilled or rejected, let's unset this alarm as the running alarm. |
| 4225 | // |
| 4226 | // NOTE: We could get here during `Actor`'s destructor, which in turn calls `Actor::Impl`'s |
| 4227 | // destructor, which destroys `runningAlarmTask`, which is us. But in this case, `actor.impl` |
| 4228 | // is already nulled out (the pointer gets nulled before the destructor runs). This is why we |
| 4229 | // captured `impl` by reference above, rather than capturing `this`. |
| 4230 | impl.maybeRunningAlarm = kj::none; |
| 4231 | })).eagerlyEvaluate([](kj::Exception&& e) { |
| 4232 | LOG_EXCEPTION("actorAlarmCleanup", e); |
| 4233 | }).fork(); |
| 4234 | co_return kj::mv(scheduledAlarm.resultFulfiller); |
| 4235 | } |
| 4236 | |
| 4237 | kj::Maybe<api::ExportedHandler&> Worker::Actor::getHandler() { |
| 4238 | KJ_SWITCH_ONEOF(impl->classInstance) { |
| 4239 | KJ_CASE_ONEOF(_, Impl::NoClass) { |
| 4240 | return kj::none; |
| 4241 | } |
| 4242 | KJ_CASE_ONEOF(_, Worker::ActorClassInfo*) { |
| 4243 | KJ_FAIL_ASSERT("ensureConstructed() wasn't called"); |
| 4244 | } |
| 4245 | KJ_CASE_ONEOF(_, Impl::Initializing) { |
| 4246 | // This shouldn't be possible because ensureConstructed() would have initiated the |
| 4247 | // construction task which would have taken an input lock as well as the isolate lock, |
| 4248 | // which should have prevented any other code from executing on the actor until they |
| 4249 | // were released. |
| 4250 | KJ_FAIL_ASSERT("actor still initializing when getHandler() called"); |
| 4251 | } |
| 4252 | KJ_CASE_ONEOF(handler, api::ExportedHandler) { |
| 4253 | return handler; |
| 4254 | } |
| 4255 | KJ_CASE_ONEOF(exception, kj::Exception) { |
| 4256 | kj::throwFatalException(exception.clone()); |
| 4257 | } |
| 4258 | } |
| 4259 | KJ_UNREACHABLE; |
| 4260 | } |
| 4261 | |
| 4262 | ActorObserver& Worker::Actor::getMetrics() { |
| 4263 | return *impl->metrics; |
| 4264 | } |
| 4265 | |
| 4266 | InputGate& Worker::Actor::getInputGate() { |
| 4267 | return impl->inputGate; |
| 4268 | } |
| 4269 | |
| 4270 | OutputGate& Worker::Actor::getOutputGate() { |
| 4271 | return impl->outputGate; |
| 4272 | } |
| 4273 | |
| 4274 | kj::Maybe<IoContext&> Worker::Actor::getIoContext() { |
| 4275 | return impl->ioContext.map([](kj::Own<IoContext>& rc) -> IoContext& { return *rc; }); |
| 4276 | } |
| 4277 | |
| 4278 | void Worker::Actor::setIoContext(kj::Own<IoContext> context) { |
| 4279 | KJ_REQUIRE(impl->ioContext == kj::none); |
| 4280 | KJ_IF_SOME(f, impl->abortFulfiller) { |
| 4281 | f.get()->fulfill(context->onAbort()); |
| 4282 | impl->abortFulfiller = kj::none; |
| 4283 | } |
| 4284 | auto& limitEnforcer = context->getLimitEnforcer(); |
| 4285 | impl->ioContext = kj::mv(context); |
| 4286 | impl->metricsFlushLoopTask = |
| 4287 | impl->metrics->flushLoop(impl->timerChannel, limitEnforcer) |
| 4288 | .eagerlyEvaluate([](kj::Exception&& e) { LOG_EXCEPTION("actorMetricsFlushLoop", e); }); |
| 4289 | } |
| 4290 | |
| 4291 | jsg::JsObject Worker::Actor::getCtx(jsg::Lock& js) { |
| 4292 | return KJ_REQUIRE_NONNULL(impl->ctxObject).getHandle(js); |
| 4293 | } |
| 4294 | |
| 4295 | jsg::JsValue Worker::Actor::getEnv(jsg::Lock& js) { |
| 4296 | return jsg::JsValue(KJ_REQUIRE_NONNULL(worker->impl->env).getHandle(js)); |
| 4297 | } |
| 4298 | |
| 4299 | kj::Maybe<Worker::Actor::HibernationManager&> Worker::Actor::getHibernationManager() { |
| 4300 | return impl->hibernationManager.map( |
| 4301 | [](kj::Own<HibernationManager>& hib) -> HibernationManager& { return *hib; }); |
| 4302 | } |
| 4303 | |
| 4304 | void Worker::Actor::setHibernationManager(kj::Own<HibernationManager> hib) { |
| 4305 | KJ_REQUIRE(impl->hibernationManager == kj::none); |
| 4306 | hib->setTimerChannel(impl->timerChannel); |
| 4307 | // Not the cleanest way to provide hibernation manager with a timer channel reference, but |
| 4308 | // where HibernationManager is constructed (actor-state), we don't have a timer channel ref. |
| 4309 | impl->hibernationManager = kj::mv(hib); |
| 4310 | } |
| 4311 | |
| 4312 | kj::Maybe<uint16_t> Worker::Actor::getHibernationEventType() { |
| 4313 | return impl->hibernationEventType; |
| 4314 | } |
| 4315 | |
| 4316 | kj::Own<Worker::Actor> Worker::Actor::addRef() { |
| 4317 | KJ_IF_SOME(t, tracker) { |
| 4318 | // We can attachToThisReference() here, attached object's lifetime being tied to refcounted |
| 4319 | // instance is deliberate. |
| 4320 | return kj::addRef(*this).attachToThisReference(t.get()->startRequest()); |
| 4321 | } else { |
| 4322 | return kj::addRef(*this); |
| 4323 | } |
| 4324 | } |
| 4325 | |
| 4326 | // ======================================================================================= |
| 4327 | |
| 4328 | uint Worker::Isolate::getCurrentLoad() const { |
| 4329 | return __atomic_load_n(&impl->lockAttemptGauge, __ATOMIC_RELAXED); |
| 4330 | } |
| 4331 | |
| 4332 | uint Worker::Isolate::getLockSuccessCount() const { |
| 4333 | return __atomic_load_n(&impl->lockSuccessCount, __ATOMIC_RELAXED); |
| 4334 | } |
| 4335 | |
| 4336 | kj::Own<const Worker::Script> Worker::Isolate::newScript(kj::StringPtr scriptId, |
| 4337 | const Script::Source& source, |
| 4338 | IsolateObserver::StartType startType, |
| 4339 | SpanParent parentSpan, |
| 4340 | kj::Own<workerd::VirtualFileSystem> vfs, |
| 4341 | bool logNewScript, |
| 4342 | kj::Maybe<ValidationErrorReporter&> errorReporter, |
| 4343 | kj::Maybe<kj::Own<api::pyodide::ArtifactBundler_State>> artifacts, |
| 4344 | kj::Maybe<kj::Arc<workerd::jsg::modules::ModuleRegistry>> maybeNewModuleRegistry) const { |
| 4345 | // Script doesn't already exist, so compile it. |
| 4346 | return kj::atomicRefcounted<Script>(kj::atomicAddRef(*this), scriptId, source, startType, |
| 4347 | logNewScript, errorReporter, kj::mv(artifacts), kj::mv(parentSpan), kj::mv(vfs), |
| 4348 | kj::mv(maybeNewModuleRegistry)); |
| 4349 | } |
| 4350 | |
| 4351 | void Worker::Isolate::completedRequest() const { |
| 4352 | limitEnforcer->completedRequest(id); |
| 4353 | } |
| 4354 | |
| 4355 | bool Worker::Isolate::isInspectorEnabled() const { |
| 4356 | return impl->inspector != kj::none; |
| 4357 | } |
| 4358 | |
| 4359 | namespace { |
| 4360 | |
| 4361 | // We only run the inspector within process sandboxes. There, it is safe to query the real clock |
| 4362 | // for some things, and we do so because we may not have a IoContext available to get |
| 4363 | // Spectre-safe time. |
| 4364 | |
| 4365 | // Monotonic time in seconds with millisecond precision. |
| 4366 | double getMonotonicTimeForProcessSandboxOnly() { |
| 4367 | KJ_REQUIRE(!isMultiTenantProcess(), "precise timing not safe in multi-tenant processes"); |
| 4368 | auto timePoint = kj::systemPreciseMonotonicClock().now(); |
| 4369 | return (timePoint - kj::origin<kj::TimePoint>()) / kj::MILLISECONDS / 1e3; |
| 4370 | } |
| 4371 | |
| 4372 | // Wall time in seconds with millisecond precision. |
| 4373 | double getWallTimeForProcessSandboxOnly() { |
| 4374 | KJ_REQUIRE(!isMultiTenantProcess(), "precise timing not safe in multi-tenant processes"); |
| 4375 | auto timePoint = kj::systemPreciseCalendarClock().now(); |
| 4376 | return (timePoint - kj::UNIX_EPOCH) / kj::MILLISECONDS / 1e3; |
| 4377 | } |
| 4378 | } // namespace |
| 4379 | |
| 4380 | class Worker::Isolate::ResponseStreamWrapper final: public kj::AsyncOutputStream { |
| 4381 | public: |
| 4382 | ResponseStreamWrapper(kj::Own<const Isolate> isolate, |
| 4383 | kj::String requestId, |
| 4384 | kj::Own<kj::AsyncOutputStream> inner, |
| 4385 | api::StreamEncoding encoding, |
| 4386 | RequestObserver& requestMetrics) |
| 4387 | : constIsolate(kj::mv(isolate)), |
| 4388 | requestId(kj::mv(requestId)), |
| 4389 | inner(kj::mv(inner)), |
| 4390 | requestMetrics(requestMetrics) { |
| 4391 | if (encoding == api::StreamEncoding::GZIP) { |
| 4392 | compStream.emplace().init<kj::GzipOutputStream>(decodedBuf, kj::GzipOutputStream::DECOMPRESS); |
| 4393 | } else if (encoding == api::StreamEncoding::BROTLI) { |
| 4394 | compStream.emplace().init<kj::BrotliOutputStream>( |
| 4395 | decodedBuf, kj::BrotliOutputStream::DECOMPRESS); |
| 4396 | } |
| 4397 | } |
| 4398 | |
| 4399 | ~ResponseStreamWrapper() noexcept(false) { |
| 4400 | // It's possible that we already have an isolate lock, in which case we |
| 4401 | // don't want to grab another one. Here, we can determine if we have a |
| 4402 | // lock by checking if there is a current IoContext, if we do then we |
| 4403 | // definitely have a current lock. |
| 4404 | // While it is possible for us to have an isolate lock without a current |
| 4405 | // IoContext, it is quite unlikely that we'd be cleaning up a |
| 4406 | // ResponseStreamWrapper in that situation, so checking for the current |
| 4407 | // IoContext should work fine. |
| 4408 | if (IoContext::hasCurrent()) { |
| 4409 | reportToInspector(); |
| 4410 | } else { |
| 4411 | // In this case we assume we don't have a lock and need to grab one. |
| 4412 | // If we continue to get warnings that we're taking the isolate lock |
| 4413 | // recursively here, that means we're cleaning these outside of the |
| 4414 | // IoContext but still have the isolate lock. In that case, we would |
| 4415 | // likely need to add an API to jsg::Lock to get the current lock |
| 4416 | // rather than relying on the IoContext. |
| 4417 | jsg::runInV8Stack([&](jsg::V8StackScope& stackScope) { |
| 4418 | Isolate::Impl::Lock recordedLock(*constIsolate, InspectorLock(requestMetrics), stackScope); |
| 4419 | reportToInspector(); |
| 4420 | }); |
| 4421 | } |
| 4422 | } |
| 4423 | |
| 4424 | kj::Promise<void> write(kj::ArrayPtr<const byte> buffer) override { |
| 4425 | reportBytes(buffer); |
| 4426 | return inner->write(buffer); |
| 4427 | } |
| 4428 | kj::Promise<void> write(kj::ArrayPtr<const kj::ArrayPtr<const byte>> pieces) override { |
| 4429 | for (auto& piece: pieces) { |
| 4430 | reportBytes(piece); |
| 4431 | } |
| 4432 | return inner->write(pieces); |
| 4433 | } |
| 4434 | void reportBytes(kj::ArrayPtr<const byte> buffer) { |
| 4435 | if (buffer.size() == 0) { |
| 4436 | return; |
| 4437 | } |
| 4438 | |
| 4439 | rawSize += buffer.size(); |
| 4440 | |
| 4441 | auto prevDecodedSize = decodedBuf.getWrittenSize(); |
| 4442 | KJ_IF_SOME(comp, compStream) { |
| 4443 | KJ_SWITCH_ONEOF(comp) { |
| 4444 | KJ_CASE_ONEOF(gzip, kj::GzipOutputStream) { |
| 4445 | // On invalid gzip discard the previously decoded body and rethrow to stop the stream. |
| 4446 | // This way we will report sizes up to this point but won't read any more invalid data. |
| 4447 | KJ_ON_SCOPE_FAILURE(decodedBuf.reset()); |
| 4448 | |
| 4449 | gzip.write(buffer); |
| 4450 | gzip.flush(); |
| 4451 | } |
| 4452 | KJ_CASE_ONEOF(brotli, kj::BrotliOutputStream) { |
| 4453 | KJ_ON_SCOPE_FAILURE(decodedBuf.reset()); |
| 4454 | |
| 4455 | brotli.write(buffer); |
| 4456 | brotli.flush(); |
| 4457 | } |
| 4458 | } |
| 4459 | } else { |
| 4460 | decodedBuf.write(buffer); |
| 4461 | } |
| 4462 | auto decodedChunkSize = decodedBuf.getWrittenSize() - prevDecodedSize; |
| 4463 | |
| 4464 | jsg::runInV8Stack([&](jsg::V8StackScope& stackScope) { |
| 4465 | Isolate::Impl::Lock recordedLock(*constIsolate, InspectorLock(requestMetrics), stackScope); |
| 4466 | auto& isolate = const_cast<Isolate&>(*constIsolate); |
| 4467 | |
| 4468 | KJ_IF_SOME(i, isolate.currentInspectorSession) { |
| 4469 | capnp::MallocMessageBuilder message; |
| 4470 | |
| 4471 | auto event = message.initRoot<cdp::Event>(); |
| 4472 | |
| 4473 | auto params = event.initNetworkDataReceived(); |
| 4474 | params.setRequestId(requestId); |
| 4475 | params.setEncodedDataLength(buffer.size()); |
| 4476 | params.setDataLength(decodedChunkSize); |
| 4477 | params.setTimestamp(getMonotonicTimeForProcessSandboxOnly()); |
| 4478 | |
| 4479 | i.sendNotification(event); |
| 4480 | } |
| 4481 | }); |
| 4482 | } |
| 4483 | |
| 4484 | // Intentionally not wrapping `tryPumpFrom` to force consumer to use `write` in a loop which, |
| 4485 | // in turn, will report each chunk to the inspector to show progress of a slow response. |
| 4486 | |
| 4487 | kj::Promise<void> whenWriteDisconnected() override { |
| 4488 | return inner->whenWriteDisconnected(); |
| 4489 | } |
| 4490 | |
| 4491 | private: |
| 4492 | using InspectorLock = InspectorChannelImpl::InspectorLock; |
| 4493 | |
| 4494 | kj::Own<const Isolate> constIsolate; |
| 4495 | kj::String requestId; |
| 4496 | kj::Own<kj::AsyncOutputStream> inner; |
| 4497 | size_t rawSize = 0; |
| 4498 | LimitedBodyWrapper decodedBuf; |
| 4499 | kj::Maybe<kj::OneOf<kj::GzipOutputStream, kj::BrotliOutputStream>> compStream; |
| 4500 | RequestObserver& requestMetrics; |
| 4501 | |
| 4502 | // Called when the wrapper is destroyed. |
| 4503 | // This should only ever be called when we are holding the isolate lock. |
| 4504 | void reportToInspector() { |
| 4505 | auto& isolate = const_cast<Isolate&>(*constIsolate); |
| 4506 | |
| 4507 | KJ_IF_SOME(i, isolate.currentInspectorSession) { |
| 4508 | capnp::MallocMessageBuilder message; |
| 4509 | |
| 4510 | auto event = message.initRoot<cdp::Event>(); |
| 4511 | |
| 4512 | auto params = event.initNetworkLoadingFinished(); |
| 4513 | params.setRequestId(requestId); |
| 4514 | params.setEncodedDataLength(rawSize); |
| 4515 | params.setTimestamp(getMonotonicTimeForProcessSandboxOnly()); |
| 4516 | auto response = params.initCfResponse(); |
| 4517 | KJ_IF_SOME(body, decodedBuf.getArray()) { |
| 4518 | response.setBase64Encoded(true); |
| 4519 | response.setBody(kj::encodeBase64(body)); |
| 4520 | } |
| 4521 | |
| 4522 | i.sendNotification(event); |
| 4523 | } |
| 4524 | } |
| 4525 | }; |
| 4526 | |
| 4527 | class Worker::Isolate::SubrequestClient final: public WorkerInterface { |
| 4528 | public: |
| 4529 | explicit SubrequestClient(kj::Own<const Isolate> isolate, |
| 4530 | kj::Own<WorkerInterface> inner, |
| 4531 | kj::HttpHeaderId contentEncodingHeaderId, |
| 4532 | RequestObserver& requestMetrics) |
| 4533 | : constIsolate(kj::mv(isolate)), |
| 4534 | inner(kj::mv(inner)), |
| 4535 | contentEncodingHeaderId(contentEncodingHeaderId), |
| 4536 | requestMetrics(kj::addRef(requestMetrics)) {} |
| 4537 | KJ_DISALLOW_COPY_AND_MOVE(SubrequestClient); |
| 4538 | kj::Promise<void> request(kj::HttpMethod method, |
| 4539 | kj::StringPtr url, |
| 4540 | const kj::HttpHeaders& headers, |
| 4541 | kj::AsyncInputStream& requestBody, |
| 4542 | kj::HttpService::Response& response) override; |
| 4543 | kj::Promise<void> connect(kj::StringPtr host, |
| 4544 | const kj::HttpHeaders& headers, |
| 4545 | kj::AsyncIoStream& connection, |
| 4546 | kj::HttpService::ConnectResponse& tunnel, |
| 4547 | kj::HttpConnectSettings settings) override; |
| 4548 | kj::Promise<void> prewarm(kj::StringPtr url) override; |
| 4549 | kj::Promise<ScheduledResult> runScheduled(kj::Date scheduledTime, kj::StringPtr cron) override; |
| 4550 | kj::Promise<AlarmResult> runAlarm(kj::Date scheduledTime, uint32_t retryCount) override; |
| 4551 | kj::Promise<CustomEvent::Result> customEvent(kj::Own<CustomEvent> event) override; |
| 4552 | |
| 4553 | private: |
| 4554 | kj::Own<const Isolate> constIsolate; |
| 4555 | kj::Own<WorkerInterface> inner; |
| 4556 | kj::HttpHeaderId contentEncodingHeaderId; |
| 4557 | kj::Own<RequestObserver> requestMetrics; |
| 4558 | }; |
| 4559 | |
| 4560 | kj::Promise<void> Worker::Isolate::SubrequestClient::request(kj::HttpMethod method, |
| 4561 | kj::StringPtr url, |
| 4562 | const kj::HttpHeaders& headers, |
| 4563 | kj::AsyncInputStream& requestBody, |
| 4564 | kj::HttpService::Response& response) { |
| 4565 | using InspectorLock = InspectorChannelImpl::InspectorLock; |
| 4566 | |
| 4567 | auto signalRequest = [this, method, urlCopy = kj::str(url), |
| 4568 | headersCopy = headers.clone()]() -> kj::Maybe<kj::String> { |
| 4569 | return jsg::runInV8Stack([&](jsg::V8StackScope& stackScope) -> kj::Maybe<kj::String> { |
| 4570 | Isolate::Impl::Lock recordedLock(*constIsolate, InspectorLock(*requestMetrics), stackScope); |
| 4571 | auto& lock = *recordedLock.lock; |
| 4572 | auto& isolate = const_cast<Isolate&>(*constIsolate); |
| 4573 | |
| 4574 | if (isolate.currentInspectorSession == kj::none) { |
| 4575 | return kj::none; |
| 4576 | } |
| 4577 | |
| 4578 | auto& i = KJ_ASSERT_NONNULL(isolate.currentInspectorSession); |
| 4579 | if (!i.isNetworkEnabled()) { |
| 4580 | return kj::none; |
| 4581 | } |
| 4582 | |
| 4583 | return lock.withinHandleScope([&] { |
| 4584 | auto requestId = kj::str(isolate.nextRequestId++); |
| 4585 | |
| 4586 | capnp::MallocMessageBuilder message; |
| 4587 | |
| 4588 | auto event = message.initRoot<cdp::Event>(); |
| 4589 | |
| 4590 | auto params = event.initNetworkRequestWillBeSent(); |
| 4591 | params.setRequestId(requestId); |
| 4592 | params.setLoaderId(""); |
| 4593 | params.setTimestamp(getMonotonicTimeForProcessSandboxOnly()); |
| 4594 | params.setWallTime(getWallTimeForProcessSandboxOnly()); |
| 4595 | params.setType(cdp::Page::ResourceType::FETCH); |
| 4596 | |
| 4597 | auto initiator = params.initInitiator(); |
| 4598 | initiator.setType(cdp::Network::Initiator::Type::SCRIPT); |
| 4599 | stackTraceToCDP(lock, initiator.initStack()); |
| 4600 | |
| 4601 | auto request = params.initRequest(); |
| 4602 | request.setUrl(urlCopy); |
| 4603 | request.setMethod(kj::str(method)); |
| 4604 | |
| 4605 | headersToCDP(headersCopy, request.initHeaders()); |
| 4606 | |
| 4607 | i.sendNotification(event); |
| 4608 | return kj::mv(requestId); |
| 4609 | }); |
| 4610 | }); |
| 4611 | }; |
| 4612 | |
| 4613 | auto signalResponse = |
| 4614 | [this](kj::String requestId, uint statusCode, kj::StringPtr statusText, |
| 4615 | const kj::HttpHeaders& headers, |
| 4616 | kj::Own<kj::AsyncOutputStream> responseBody) -> kj::Own<kj::AsyncOutputStream> { |
| 4617 | // Note that we cannot take the isolate lock here, because if this is a worker-to-worker |
| 4618 | // subrequest, the destination isolate's lock may already be held, and we can't take multiple |
| 4619 | // isolate locks at once as this could lead to deadlock if the lock orders aren't consistent. |
| 4620 | // |
| 4621 | // Meanwhile, though, `statusText` and `headers` may point to things that will go away |
| 4622 | // immediately after we return. So, let's construct our message now, so that we don't have to |
| 4623 | // make redundant copies. |
| 4624 | // |
| 4625 | // Note that signalResponse() is only called at all if signalRequest() determined that network |
| 4626 | // inspection is enabled. |
| 4627 | |
| 4628 | auto message = kj::heap<capnp::MallocMessageBuilder>(); |
| 4629 | |
| 4630 | auto event = message->initRoot<cdp::Event>(); |
| 4631 | |
| 4632 | auto params = event.initNetworkResponseReceived(); |
| 4633 | params.setRequestId(requestId); |
| 4634 | params.setTimestamp(getMonotonicTimeForProcessSandboxOnly()); |
| 4635 | params.setType(cdp::Page::ResourceType::OTHER); |
| 4636 | |
| 4637 | auto response = params.initResponse(); |
| 4638 | response.setStatus(statusCode); |
| 4639 | response.setStatusText(statusText); |
| 4640 | response.setProtocol("http/1.1"); |
| 4641 | KJ_IF_SOME(type, headers.get(kj::HttpHeaderId::CONTENT_TYPE)) { |
| 4642 | KJ_IF_SOME(parsed, MimeType::tryParse(type, MimeType::IGNORE_PARAMS)) { |
| 4643 | response.setMimeType(parsed.toString()); |
| 4644 | |
| 4645 | // Normally Chrome would know what it's loading based on an element or API used for |
| 4646 | // the request. We don't have that privilege, but still want network filters to work, |
| 4647 | // so we do our best-effort guess of the resource type based on its mime type. |
| 4648 | if (MimeType::HTML == parsed || MimeType::XHTML == parsed) { |
| 4649 | params.setType(cdp::Page::ResourceType::DOCUMENT); |
| 4650 | } else if (MimeType::CSS == parsed) { |
| 4651 | params.setType(cdp::Page::ResourceType::STYLESHEET); |
| 4652 | } else if (MimeType::isJavascript(parsed)) { |
| 4653 | params.setType(cdp::Page::ResourceType::SCRIPT); |
| 4654 | } else if (MimeType::isImage(parsed)) { |
| 4655 | params.setType(cdp::Page::ResourceType::IMAGE); |
| 4656 | } else if (MimeType::isAudio(parsed) || MimeType::isVideo(parsed)) { |
| 4657 | params.setType(cdp::Page::ResourceType::MEDIA); |
| 4658 | } else if (MimeType::isFont(parsed)) { |
| 4659 | params.setType(cdp::Page::ResourceType::FONT); |
| 4660 | } else if (MimeType::MANIFEST_JSON == parsed) { |
| 4661 | params.setType(cdp::Page::ResourceType::MANIFEST); |
| 4662 | } else if (MimeType::VTT == parsed) { |
| 4663 | params.setType(cdp::Page::ResourceType::TEXT_TRACK); |
| 4664 | } else if (MimeType::EVENT_STREAM == parsed) { |
| 4665 | params.setType(cdp::Page::ResourceType::EVENT_SOURCE); |
| 4666 | } else if (MimeType::isXml(parsed) || MimeType::isJson(parsed)) { |
| 4667 | params.setType(cdp::Page::ResourceType::XHR); |
| 4668 | } |
| 4669 | |
| 4670 | } else { |
| 4671 | response.setMimeType(MimeType::PLAINTEXT_STRING); |
| 4672 | } |
| 4673 | } else { |
| 4674 | response.setMimeType(MimeType::PLAINTEXT_STRING); |
| 4675 | } |
| 4676 | headersToCDP(headers, response.initHeaders()); |
| 4677 | |
| 4678 | auto encoding = api::StreamEncoding::IDENTITY; |
| 4679 | KJ_IF_SOME(encodingStr, headers.get(contentEncodingHeaderId)) { |
| 4680 | if (encodingStr == "gzip") { |
| 4681 | encoding = api::StreamEncoding::GZIP; |
| 4682 | } else if (encodingStr == "br") { |
| 4683 | encoding = api::StreamEncoding::BROTLI; |
| 4684 | } |
| 4685 | } |
| 4686 | |
| 4687 | // Defer to a later turn of the event loop so that it's safe to take a lock. |
| 4688 | return kj::newPromisedStream(kj::evalLater( |
| 4689 | [this, responseBody = kj::mv(responseBody), message = kj::mv(message), event, encoding, |
| 4690 | requestId = kj::mv(requestId)]() mutable -> kj::Own<kj::AsyncOutputStream> { |
| 4691 | // Now we know we can lock... |
| 4692 | return jsg::runInV8Stack( |
| 4693 | [&](jsg::V8StackScope& stackScope) mutable -> kj::Own<kj::AsyncOutputStream> { |
| 4694 | Isolate::Impl::Lock recordedLock(*constIsolate, InspectorLock(*requestMetrics), stackScope); |
| 4695 | auto& isolate = const_cast<Isolate&>(*constIsolate); |
| 4696 | |
| 4697 | // We shouldn't even get here if network inspection isn't active since signalRequest() would |
| 4698 | // have returned null... but double-check anyway. |
| 4699 | if (isolate.currentInspectorSession == kj::none) { |
| 4700 | return kj::mv(responseBody); |
| 4701 | } |
| 4702 | |
| 4703 | auto& i = KJ_ASSERT_NONNULL(isolate.currentInspectorSession); |
| 4704 | if (!i.isNetworkEnabled()) { |
| 4705 | return kj::mv(responseBody); |
| 4706 | } |
| 4707 | |
| 4708 | i.sendNotification(event); |
| 4709 | |
| 4710 | return kj::heap<ResponseStreamWrapper>(kj::atomicAddRef(*constIsolate), kj::mv(requestId), |
| 4711 | kj::mv(responseBody), encoding, *requestMetrics); |
| 4712 | }); |
| 4713 | })); |
| 4714 | }; |
| 4715 | using SignalResponse = decltype(signalResponse); |
| 4716 | |
| 4717 | class ResponseWrapper final: public kj::HttpService::Response { |
| 4718 | public: |
| 4719 | ResponseWrapper( |
| 4720 | kj::HttpService::Response& inner, kj::String requestId, SignalResponse signalResponse) |
| 4721 | : inner(inner), |
| 4722 | requestId(kj::mv(requestId)), |
| 4723 | signalResponse(kj::mv(signalResponse)) {} |
| 4724 | |
| 4725 | kj::Own<kj::AsyncOutputStream> send(uint statusCode, |
| 4726 | kj::StringPtr statusText, |
| 4727 | const kj::HttpHeaders& headers, |
| 4728 | kj::Maybe<uint64_t> expectedBodySize = kj::none) override { |
| 4729 | auto body = inner.send(statusCode, statusText, headers, expectedBodySize); |
| 4730 | return signalResponse(kj::mv(requestId), statusCode, statusText, headers, kj::mv(body)); |
| 4731 | } |
| 4732 | |
| 4733 | kj::Own<kj::WebSocket> acceptWebSocket(const kj::HttpHeaders& headers) override { |
| 4734 | auto webSocket = inner.acceptWebSocket(headers); |
| 4735 | // TODO(someday): Support sending WebSocket frames over CDP. For now we fake an empty |
| 4736 | // response. |
| 4737 | signalResponse(kj::mv(requestId), 101, "Switching Protocols", headers, newNullOutputStream()); |
| 4738 | return kj::mv(webSocket); |
| 4739 | } |
| 4740 | |
| 4741 | private: |
| 4742 | kj::HttpService::Response& inner; |
| 4743 | kj::String requestId; |
| 4744 | SignalResponse signalResponse; |
| 4745 | }; |
| 4746 | |
| 4747 | // For accurate lock metrics, we want to avoid taking a recursive isolate lock, so we postpone |
| 4748 | // the request until a later turn of the event loop. |
| 4749 | auto maybeRequestId = co_await kj::evalLater(kj::mv(signalRequest)); |
| 4750 | |
| 4751 | // While we checked above that the headers are valid, let's check again |
| 4752 | // after the co_await... |
| 4753 | KJ_IF_SOME(rid, maybeRequestId) { |
| 4754 | ResponseWrapper wrapper(response, kj::mv(rid), kj::mv(signalResponse)); |
| 4755 | co_await inner->request(method, url, headers, requestBody, wrapper); |
| 4756 | } else { |
| 4757 | co_await inner->request(method, url, headers, requestBody, response); |
| 4758 | } |
| 4759 | } |
| 4760 | |
| 4761 | kj::Promise<void> Worker::Isolate::SubrequestClient::connect(kj::StringPtr host, |
| 4762 | const kj::HttpHeaders& headers, |
| 4763 | kj::AsyncIoStream& connection, |
| 4764 | kj::HttpService::ConnectResponse& tunnel, |
| 4765 | kj::HttpConnectSettings settings) { |
| 4766 | // TODO(someday): EW-7116 Figure out how to represent TCP connections in the devtools network tab. |
| 4767 | return inner->connect(host, headers, connection, tunnel, kj::mv(settings)); |
| 4768 | } |
| 4769 | |
| 4770 | // TODO(someday): Log other kinds of subrequests? |
| 4771 | kj::Promise<void> Worker::Isolate::SubrequestClient::prewarm(kj::StringPtr url) { |
| 4772 | return inner->prewarm(url); |
| 4773 | } |
| 4774 | kj::Promise<WorkerInterface::ScheduledResult> Worker::Isolate::SubrequestClient::runScheduled( |
| 4775 | kj::Date scheduledTime, kj::StringPtr cron) { |
| 4776 | return inner->runScheduled(scheduledTime, cron); |
| 4777 | } |
| 4778 | kj::Promise<WorkerInterface::AlarmResult> Worker::Isolate::SubrequestClient::runAlarm( |
| 4779 | kj::Date scheduledTime, uint32_t retryCount) { |
| 4780 | return inner->runAlarm(scheduledTime, retryCount); |
| 4781 | } |
| 4782 | kj::Promise<WorkerInterface::CustomEvent::Result> Worker::Isolate::SubrequestClient::customEvent( |
| 4783 | kj::Own<CustomEvent> event) { |
| 4784 | return inner->customEvent(kj::mv(event)); |
| 4785 | } |
| 4786 | |
| 4787 | kj::Own<WorkerInterface> Worker::Isolate::wrapSubrequestClient(kj::Own<WorkerInterface> client, |
| 4788 | kj::HttpHeaderId contentEncodingHeaderId, |
| 4789 | RequestObserver& requestMetrics) const { |
| 4790 | if (impl->inspector != kj::none) { |
| 4791 | client = kj::heap<SubrequestClient>( |
| 4792 | kj::atomicAddRef(*this), kj::mv(client), contentEncodingHeaderId, requestMetrics); |
| 4793 | } |
| 4794 | |
| 4795 | return client; |
| 4796 | } |
| 4797 | |
| 4798 | } // namespace workerd |