firefox-main/ipc/glue
Name Description Size Coverage
AsyncBlockers.h AsyncBlockers provide a simple registration service that allows to suspend completion of a particular task until all registered entries have been cleared. This can be used to implement a similar service to nsAsyncShutdownService in processes where it wouldn't normally be available. This class is thread-safe. 2747 -
BackgroundChild.h 2341 -
BackgroundChildImpl.cpp 13282 -
BackgroundChildImpl.h 5792 -
BackgroundImpl.cpp 39624 -
BackgroundParent.h 3454 -
BackgroundParentImpl.cpp 48792 -
BackgroundParentImpl.h 14329 -
BackgroundStarterChild.h 1141 -
BackgroundStarterParent.h 1548 -
BackgroundUtils.cpp 44513 -
BackgroundUtils.h Convert a PrincipalInfo to an nsIPrincipal. MUST be called on the main thread. 5991 -
BigBuffer.cpp 3590 -
BigBuffer.h 4203 -
ByteBuf.h A type that can be sent without needing to make a copy during serialization. In addition the receiver can take ownership of the data to avoid having to make an additional copy. 1999 -
ByteBufUtils.h A type that can be sent without needing to make a copy during serialization. In addition the receiver can take ownership of the data to avoid having to make an additional copy. 1985 -
components.conf 718 -
CrashReporterClient.cpp static 2156 -
CrashReporterClient.h 1310 -
CrashReporterHelper.h This class encapsulates the common elements of crash report handling for toplevel protocols representing processes. To use this class, you should: 1. Declare a method to initialize the crash reporter in your IPDL: `async InitCrashReporter(NativeThreadId threadId)` 2. Inherit from this class with the name of your derived class as the type parameter. Ex: `class MyClass : public CrashReporterHelper<MyClass>` 3. Provide a public `PROCESS_TYPE` constant for your class. Ex: ``` public: static constexpr GeckoProcessType PROCESS_TYPE = GeckoProcessType_GMPlugin; ``` 4. When your protocol actor is destroyed with a reason of `AbnormalShutdown`, you should call `GenerateCrashReport()`. If you need the crash report ID it will be copied in the second optional parameter upon successful crash report generation. 3745 -
CrashReporterHost.cpp static 7230 -
CrashReporterHost.h 5006 -
CrossProcessMutex.h CrossProcessMutex @param name A name which can reference this lock (currently unused) 3378 -
CrossProcessMutex_posix.cpp 2960 -
CrossProcessMutex_unimplemented.cpp 1236 -
CrossProcessMutex_windows.cpp 1909 -
CrossProcessSemaphore.h CrossProcessSemaphore @param name A name which can reference this lock (currently unused) 3058 -
CrossProcessSemaphore_mach.cpp static 2890 -
CrossProcessSemaphore_posix.cpp static 3726 -
CrossProcessSemaphore_unimplemented.cpp static 1792 -
CrossProcessSemaphore_windows.cpp static 2430 -
DataPipe.cpp 27904 -
DataPipe.h aReceiverSide 6754 -
EagerIPCStream.h 627 -
Endpoint.cpp mOtherSide 7430 -
Endpoint.h An endpoint represents one end of a partially initialized IPDL channel. To set up a new top-level protocol: Endpoint<PFooParent> parentEp; Endpoint<PFooChild> childEp; nsresult rv; rv = PFoo::CreateEndpoints(&parentEp, &childEp); Endpoints can be passed in IPDL messages or sent to other threads using PostTask. Once an Endpoint has arrived at its destination process and thread, you need to create the top-level actor and bind it to the endpoint: FooParent* parent = new FooParent(); bool rv1 = parentEp.Bind(parent, processActor); bool rv2 = parent->SendBar(...); (See Bind below for an explanation of processActor.) Once the actor is bound to the endpoint, it can send and receive messages. If creating endpoints for a [NeedsOtherPid] actor, you're required to also pass in parentPid and childPid, which are the pids of the processes in which the parent and child endpoints will be used. 10596 -
EnumSerializer.h Generic enum serializer. Consider using the specializations below, such as ContiguousEnumSerializer. This is a generic serializer for any enum type used in IPDL. Programmers can define ParamTraits<E> for enum type E by deriving EnumSerializer<E, MyEnumValidator> where MyEnumValidator is a struct that has to define a static IsLegalValue function returning whether a given value is a legal value of the enum type at hand. \sa https://developer.mozilla.org/en/IPDL/Type_Serialization 6025 -
EnvironmentMap.h 2207 -
ExtensionKitUtils.h 2618 -
ExtensionKitUtils.mm 4758 -
FileDescriptor.cpp 2446 -
FileDescriptor.h 2363 -
FileDescriptorShuffle.cpp 3649 -
FileDescriptorShuffle.h 2210 -
FileDescriptorUtils.cpp 2581 -
FileDescriptorUtils.h 1490 -
ForkServer.cpp Preload any resources that the forked child processes might need, and which might change incompatibly or become unavailable by the time they're started. For example: the omnijar files, or certain shared libraries. 13185 -
ForkServer.h 1138 -
ForkServiceChild.cpp 9523 -
ForkServiceChild.h This is the interface to the fork server. When the chrome process calls |ForkServiceChild| to create a new process, this class send a message to the fork server through a pipe and get the PID of the new process from the reply. 3952 -
GeckoChildProcessHost.cpp 69456 -
GeckoChildProcessHost.h 11838 -
IdleSchedulerChild.cpp 4433 -
IdleSchedulerChild.h 2058 -
IdleSchedulerParent.cpp 15395 -
IdleSchedulerParent.h 4100 -
InputStreamParams.ipdlh 1963 -
InputStreamUtils.cpp 7053 -
InputStreamUtils.h 1925 -
IOThread.cpp static 3311 -
IOThread.h 2776 -
IPCCore.h 512 -
IPCForwards.h 1091 -
IPCMessageUtils.h A helper class for serializing empty structs. Since the struct is empty there is nothing to write, and a priori we know the result of the read. 9748 -
IPCMessageUtilsSpecializations.cpp static 1896 -
IPCMessageUtilsSpecializations.h 26777 -
IPCStream.ipdlh 679 -
IPCStreamUtils.cpp 6437 -
IPCStreamUtils.h 1937 -
IPCTypes.h 510 -
IPDLStructMember.h 1181 -
LaunchError.h 2589 -
MessageChannel.cpp IPC design: There are two kinds of messages: async and sync. Sync messages are blocking. Terminology: To dispatch a message Foo is to run the RecvFoo code for it. This is also called "handling" the message. Sync and async messages can sometimes "nest" inside other sync messages (i.e., while waiting for the sync reply, we can dispatch the inner message). The three possible nesting levels are NOT_NESTED, NESTED_INSIDE_SYNC, and NESTED_INSIDE_CPOW. The intended uses are: NOT_NESTED - most messages. NESTED_INSIDE_SYNC - CPOW-related messages, which are always sync and can go in either direction. NESTED_INSIDE_CPOW - messages where we don't want to dispatch incoming CPOWs while waiting for the response. These nesting levels are ordered: NOT_NESTED, NESTED_INSIDE_SYNC, NESTED_INSIDE_CPOW. Async messages cannot be NESTED_INSIDE_SYNC but they can be NESTED_INSIDE_CPOW. To avoid jank, the parent process is not allowed to send NOT_NESTED sync messages. When a process is waiting for a response to a sync message M0, it will dispatch an incoming message M if: 1. M has a higher nesting level than M0, or 2. if M has the same nesting level as M0 and we're in the child, or 3. if M has the same nesting level as M0 and it was sent by the other side while dispatching M0. The idea is that messages with higher nesting should take precendence. The purpose of rule 2 is to handle a race where both processes send to each other simultaneously. In this case, we resolve the race in favor of the parent (so the child dispatches first). Messages satisfy the following properties: A. When waiting for a response to a sync message, we won't dispatch any messages of a lower nesting level. B. Messages of the same nesting level will be dispatched roughly in the order they were sent. The exception is when the parent and child send sync messages to each other simulataneously. In this case, the parent's message is dispatched first. While it is dispatched, the child may send further nested messages, and these messages may be dispatched before the child's original message. We can consider ordering to be preserved here because we pretend that the child's original message wasn't sent until after the parent's message is finished being dispatched. When waiting for a sync message reply, we dispatch an async message only if it is NESTED_INSIDE_CPOW. Normally NESTED_INSIDE_CPOW async messages are sent only from the child. However, the parent can send NESTED_INSIDE_CPOW async messages when it is creating a bridged protocol. 84951 -
MessageChannel.h This sends a special message that is processed on the IO thread, so that other actors can know that the process will soon shutdown. 29900 -
MessageLink.cpp 6205 -
MessageLink.h 2649 -
MessagePump.cpp namespace ipc 9814 -
MessagePump.h ` The MessagePumpForAndroidUI exists to enable IPDL in the Android UI thread. The Android UI thread event loop is controlled by Android. This prevents running an existing MessagePump implementation in the Android UI thread. In order to enable IPDL on the Android UI thread it is necessary to have a non-looping MessagePump. This class enables forwarding of nsIRunnables from MessageLoop::PostTask_Helper to the registered nsIEventTarget with out the need to control the event loop. The only member function that should be invoked is GetXPCOMThread. All other member functions will invoke MOZ_CRASH 5653 -
MessagePump_android.cpp 800 -
MessagePump_mac.mm 3155 -
MessagePump_windows.cpp 2728 -
MiniTransceiver.cpp Initialize the IO vector for sending data and the control buffer for sending FDs. 7657 -
MiniTransceiver.h This simple implementation handles the transmissions of IPC messages. It works according to a strict request-response paradigm, no concurrent messaging is allowed. Sending a message from A to B must be followed by another one from B to A. Because of this we don't need to handle data crossing the boundaries of a message. Transmission is done via blocking I/O to avoid the complexity of asynchronous I/O. 3520 -
moz.build 8229 -
Neutering.h This header declares RAII wrappers for Window neutering. See WindowsMessageLoop.cpp for more details. 1728 -
NodeChannel.cpp 9732 -
NodeChannel.h 6332 -
NodeController.cpp static 32648 -
NodeController.h 7370 -
nsIIPCSerializableInputStream.h 4804 -
PBackground.ipdl Issue an asynchronous request that will be used in a synchronous fashion through complex machinations described in `PBackgroundLSRequest.ipdl` and `LSObject.h`. 10197 -
PBackgroundSharedTypes.ipdlh 2689 -
PBackgroundStarter.ipdl 461 -
PBackgroundTest.ipdl 551 -
PIdleScheduler.ipdl PIdleScheduler is the protocol for cross-process idle scheduling. Only child processes participate in the scheduling and parent process can run its idle tasks whenever it needs to. The scheduler keeps track of the following things. - Activity of the main thread of each child process. A process is active when it is running tasks. Because of performance cross-process counters in shared memory are used for the activity tracking. There is one counter counting the activity state of all the processes and one counter for each process. This way if a child process crashes, the global counter can be updated by decrementing the per process counter from it. - Child processes running prioritized operation. Top level page loads is an example of a prioritized operation. When such is ongoing, idle tasks are less likely to run. - Idle requests. When a child process locally has idle tasks to run, it requests idle time from the scheduler. Initially requests go to a wait list and the scheduler runs and if there are free logical cores for the child processes, idle time is given to the child process, and the process goes to the idle list. Once idle time has been consumed or there are no tasks to process, child process informs the scheduler and the process is moved back to the default queue. 2852 -
ProcessChild.cpp static 4290 -
ProcessChild.h Exit *now*. Do not shut down XPCOM, do not pass Go, do not run static destructors, do not collect $200. 2264 -
ProcessUtils.h 2490 -
ProcessUtils_bsd.cpp 582 -
ProcessUtils_common.cpp 4733 -
ProcessUtils_linux.cpp 469 -
ProcessUtils_mac.mm 3947 -
ProcessUtils_none.cpp 379 -
ProtocolMessageUtils.h 3282 -
ProtocolTypes.ipdlh 363 -
ProtocolUtils.cpp static 30914 -
ProtocolUtils.h 31163 -
PUtilityMediaService.ipdl 1597 -
PUtilityProcess.ipdl This method is used to notify a child process to start processing module loading events in UntrustedModulesProcessor. This should be called when the parent process has gone idle. 5093 -
RandomAccessStreamParams.ipdlh 857 -
RandomAccessStreamUtils.cpp 2624 -
RandomAccessStreamUtils.h 1440 -
RustMessageUtils.h TODO: Should be able to initialize `result` in-place instead 2646 -
ScopedPort.cpp 2017 -
ScopedPort.h 2194 -
SerializedStructuredCloneBuffer.cpp 3002 -
SerializedStructuredCloneBuffer.h IPC_GLUE_SERIALIZEDSTRUCTUREDCLONEBUFFER_H_ 2267 -
SerializeToBytesUtil.h 2139 -
SetProcessTitle.cpp 1098 -
SetProcessTitle.h 516 -
SharedMemoryCursor.cpp aWriteToShmem 3419 -
SharedMemoryCursor.h 3638 -
SharedMemoryHandle.cpp 6872 -
SharedMemoryHandle.h The size of the shared memory region to which this handle refers. 8379 -
SharedMemoryMapping.cpp aOffset 6133 -
SharedMemoryMapping.h A leaked memory mapping. This memory will never be unmapped. 9140 -
SharedMemoryPlatform.h Functions that need to be implemented for each platform. These are static methods of a class to simplify access (the class can be made a friend to give access to platform implementations). 3998 -
SharedMemoryPlatform_android.cpp 3906 -
SharedMemoryPlatform_mach.cpp nothing 7196 -
SharedMemoryPlatform_posix.cpp 15673 -
SharedMemoryPlatform_windows.cpp 8106 -
Shmem.cpp unused 6184 -
Shmem.h |Shmem| is one agent in the IPDL shared memory scheme. The way it works is essentially (1) C++ code calls, say, |parentActor->AllocShmem(size)| (2) IPDL-generated code creates a |mozilla::ipc::SharedMemoryMapping| wrapping the bare OS shmem primitives. The code then adds the new SharedMemory to the set of shmem segments being managed by IPDL. (3) IPDL-generated code "shares" the new SharedMemory to the child process, and then sends a special asynchronous IPC message to the child notifying it of the creation of the segment. (What this means is OS specific.) (4a) The child receives the special IPC message, and using the |MutableSharedMemoryHandle| it was passed, creates a |SharedMemoryMapping| in the child process. (4b) After sending the "shmem-created" IPC message, IPDL-generated code in the parent returns a |mozilla::ipc::Shmem| back to the C++ caller of |parentActor->AllocShmem()|. The |Shmem| is a "weak reference" to the underlying |SharedMemory|, which is managed by IPDL-generated code. C++ consumers of |Shmem| can't get at the underlying |SharedMemoryMapping|. If parent code wants to give access rights to the Shmem to the child, it does so by sending its |Shmem| to the child, in an IPDL message. The parent's |Shmem| then "dies", i.e. becomes inaccessible. This process could be compared to passing a "shmem-access baton" between parent and child. 6082 -
ShmemMessageUtils.h 679 -
SideVariant.h Helper type used by IPDL structs and unions to hold actor pointers with a dynamic side. When sent over IPC, ParentSide will be used for send/recv on parent actors, and ChildSide will be used for send/recv on child actors. 5928 -
StringUtil.cpp 2489 -
TaintingIPCUtils.h 881 -
TaskFactory.h This is based on the ScopedRunnableMethodFactory from ipc/chromium/src/base/task.h Chromium's factories assert if tasks are created and run on different threads, which is something we need to do in PluginModuleParent (hang UI vs. main thread). TaskFactory just provides cancellable tasks that don't assert this. This version also allows both ScopedMethod and regular Tasks to be generated by the same Factory object. 2760 -
test -
TransportSecurityInfoUtils.cpp 1806 -
TransportSecurityInfoUtils.h 1699 -
URIParams.ipdlh 2123 -
URIUtils.cpp 3446 -
URIUtils.h 1325 -
UtilityMediaService.cpp 1236 -
UtilityMediaService.h 675 -
UtilityMediaServiceChild.cpp static 8810 -
UtilityMediaServiceChild.h 3331 -
UtilityMediaServiceParent.cpp static 6964 -
UtilityMediaServiceParent.h 1907 -
UtilityProcessChild.cpp static 14039 -
UtilityProcessChild.h 4344 -
UtilityProcessHost.cpp remoteType 12517 -
UtilityProcessHost.h 5811 -
UtilityProcessImpl.cpp static 4413 -
UtilityProcessImpl.h 1217 -
UtilityProcessManager.cpp isLocked 24959 -
UtilityProcessManager.h 7856 -
UtilityProcessParent.cpp 6949 -
UtilityProcessParent.h 2723 -
UtilityProcessSandboxing.cpp 2401 -
UtilityProcessSandboxing.h 1136 -
WindowsMessageLoop.cpp The Windows-only code below exists to solve a general problem with deadlocks that we experience when sending synchronous IPC messages to processes that contain native windows (i.e. HWNDs). Windows (the OS) sends synchronous messages between parent and child HWNDs in multiple circumstances (e.g. WM_PARENTNOTIFY, WM_NCACTIVATE, etc.), even when those HWNDs are controlled by different threads or different processes. Thus we can very easily end up in a deadlock by a call stack like the following: Process A: - CreateWindow(...) creates a "parent" HWND. - SendCreateChildWidget(HWND) is a sync IPC message that sends the "parent" HWND over to Process B. Process A blocks until a response is received from Process B. Process B: - RecvCreateWidget(HWND) gets the "parent" HWND from Process A. - CreateWindow(..., HWND) creates a "child" HWND with the parent from process A. - Windows (the OS) generates a WM_PARENTNOTIFY message that is sent synchronously to Process A. Process B blocks until a response is received from Process A. Process A, however, is blocked and cannot process the message. Both processes are deadlocked. The example above has a few different workarounds (e.g. setting the WS_EX_NOPARENTNOTIFY style on the child window) but the general problem is persists. Once two HWNDs are parented we must not block their owning threads when manipulating either HWND. Windows requires any application that hosts native HWNDs to always process messages or risk deadlock. Given our architecture the only way to meet Windows' requirement and allow for synchronous IPC messages is to pump a miniature message loop during a sync IPC call. We avoid processing any queued messages during the loop (with one exception, see below), but "nonqueued" messages (see http://msdn.microsoft.com/en-us/library/ms644927(VS.85).aspx under the section "Nonqueued messages") cannot be avoided. Those messages are trapped in a special window procedure where we can either ignore the message or process it in some fashion. Queued and "non-queued" messages will be processed during Interrupt calls if modal UI related api calls block an Interrupt in-call in the child. To prevent windows from freezing, and to allow concurrent processing of critical events (such as painting), we spin a native event dispatch loop while these in-calls are blocked. 36993 -
WindowsMessageLoop.h namespace windows 3367 -