| /* |
| * |
| * Copyright (c) 2020-2021 Project CHIP Authors |
| * All rights reserved. |
| * |
| * Licensed under the Apache License, Version 2.0 (the "License"); |
| * you may not use this file except in compliance with the License. |
| * You may obtain a copy of the License at |
| * |
| * http://www.apache.org/licenses/LICENSE-2.0 |
| * |
| * Unless required by applicable law or agreed to in writing, software |
| * distributed under the License is distributed on an "AS IS" BASIS, |
| * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| * See the License for the specific language governing permissions and |
| * limitations under the License. |
| */ |
| |
| /** |
| * @file |
| * This file defines objects for a CHIP Interaction Data model Engine which handle unsolicited IM message, and |
| * manage different kinds of IM client and handlers. |
| * |
| */ |
| |
| #include "InteractionModelEngine.h" |
| |
| #include <cinttypes> |
| |
| #include "access/RequestPath.h" |
| #include "access/SubjectDescriptor.h" |
| #include <app/RequiredPrivilege.h> |
| #include <lib/core/CHIPTLVUtilities.hpp> |
| #include <lib/support/CodeUtils.h> |
| |
| extern bool emberAfContainsAttribute(chip::EndpointId endpoint, chip::ClusterId clusterId, chip::AttributeId attributeId); |
| |
| namespace chip { |
| namespace app { |
| |
| using Protocols::InteractionModel::Status; |
| |
| InteractionModelEngine sInteractionModelEngine; |
| |
| InteractionModelEngine::InteractionModelEngine() {} |
| |
| InteractionModelEngine * InteractionModelEngine::GetInstance() |
| { |
| return &sInteractionModelEngine; |
| } |
| |
| CHIP_ERROR InteractionModelEngine::Init(Messaging::ExchangeManager * apExchangeMgr, FabricTable * apFabricTable, |
| CASESessionManager * apCASESessionMgr) |
| { |
| VerifyOrReturnError(apFabricTable != nullptr, CHIP_ERROR_INVALID_ARGUMENT); |
| VerifyOrReturnError(apExchangeMgr != nullptr, CHIP_ERROR_INVALID_ARGUMENT); |
| |
| mpExchangeMgr = apExchangeMgr; |
| mpFabricTable = apFabricTable; |
| mpCASESessionMgr = apCASESessionMgr; |
| |
| ReturnErrorOnFailure(mpFabricTable->AddFabricDelegate(this)); |
| ReturnErrorOnFailure(mpExchangeMgr->RegisterUnsolicitedMessageHandlerForProtocol(Protocols::InteractionModel::Id, this)); |
| |
| mReportingEngine.Init(); |
| mMagic++; |
| |
| StatusIB::RegisterErrorFormatter(); |
| |
| return CHIP_NO_ERROR; |
| } |
| |
| void InteractionModelEngine::Shutdown() |
| { |
| CommandHandlerInterface * handlerIter = mCommandHandlerList; |
| |
| // |
| // Walk our list of command handlers and de-register them, before finally |
| // nulling out the list entirely. |
| // |
| while (handlerIter) |
| { |
| CommandHandlerInterface * nextHandler = handlerIter->GetNext(); |
| handlerIter->SetNext(nullptr); |
| handlerIter = nextHandler; |
| } |
| |
| mCommandHandlerList = nullptr; |
| |
| // Increase magic number to invalidate all Handle-s. |
| mMagic++; |
| |
| mCommandHandlerObjs.ReleaseAll(); |
| |
| mTimedHandlers.ForEachActiveObject([this](TimedHandler * obj) -> Loop { |
| mpExchangeMgr->CloseAllContextsForDelegate(obj); |
| return Loop::Continue; |
| }); |
| |
| mTimedHandlers.ReleaseAll(); |
| |
| mReadHandlers.ReleaseAll(); |
| |
| // Shut down any subscription clients that are still around. They won't be |
| // able to work after this point anyway, since we're about to drop our refs |
| // to them. |
| ShutdownAllSubscriptions(); |
| |
| // |
| // We hold weak references to ReadClient objects. The application ultimately |
| // actually owns them, so it's on them to eventually shut them down and free them |
| // up. |
| // |
| // However, we should null out their pointers back to us at the very least so that |
| // at destruction time, they won't attempt to reach back here to remove themselves |
| // from this list. |
| // |
| for (auto * readClient = mpActiveReadClientList; readClient != nullptr;) |
| { |
| readClient->mpImEngine = nullptr; |
| auto * tmpClient = readClient->GetNextClient(); |
| readClient->SetNextClient(nullptr); |
| readClient = tmpClient; |
| } |
| |
| // |
| // After that, we just null out our tracker. |
| // |
| mpActiveReadClientList = nullptr; |
| |
| for (auto & writeHandler : mWriteHandlers) |
| { |
| writeHandler.Abort(); |
| } |
| |
| mReportingEngine.Shutdown(); |
| mAttributePathPool.ReleaseAll(); |
| mEventPathPool.ReleaseAll(); |
| mDataVersionFilterPool.ReleaseAll(); |
| mpExchangeMgr->UnregisterUnsolicitedMessageHandlerForProtocol(Protocols::InteractionModel::Id); |
| |
| mpCASESessionMgr = nullptr; |
| |
| // |
| // We _should_ be clearing these out, but doing so invites a world |
| // of trouble. #21233 tracks fixing the underlying assumptions to make |
| // this possible. |
| // |
| // mpFabricTable = nullptr; |
| // mpExchangeMgr = nullptr; |
| } |
| |
| uint32_t InteractionModelEngine::GetNumActiveReadHandlers() const |
| { |
| return static_cast<uint32_t>(mReadHandlers.Allocated()); |
| } |
| |
| uint32_t InteractionModelEngine::GetNumActiveReadHandlers(ReadHandler::InteractionType aType) const |
| { |
| uint32_t count = 0; |
| |
| mReadHandlers.ForEachActiveObject([aType, &count](const ReadHandler * handler) { |
| if (handler->IsType(aType)) |
| { |
| count++; |
| } |
| |
| return Loop::Continue; |
| }); |
| |
| return count; |
| } |
| |
| uint32_t InteractionModelEngine::GetNumActiveReadHandlers(ReadHandler::InteractionType aType, FabricIndex aFabricIndex) const |
| { |
| uint32_t count = 0; |
| |
| mReadHandlers.ForEachActiveObject([aType, aFabricIndex, &count](const ReadHandler * handler) { |
| if (handler->IsType(aType) && handler->GetAccessingFabricIndex() == aFabricIndex) |
| { |
| count++; |
| } |
| |
| return Loop::Continue; |
| }); |
| |
| return count; |
| } |
| |
| ReadHandler * InteractionModelEngine::ActiveHandlerAt(unsigned int aIndex) |
| { |
| if (aIndex >= mReadHandlers.Allocated()) |
| { |
| return nullptr; |
| } |
| |
| unsigned int i = 0; |
| ReadHandler * ret = nullptr; |
| |
| mReadHandlers.ForEachActiveObject([aIndex, &i, &ret](ReadHandler * handler) { |
| if (i == aIndex) |
| { |
| ret = handler; |
| return Loop::Break; |
| } |
| |
| i++; |
| return Loop::Continue; |
| }); |
| |
| return ret; |
| } |
| |
| WriteHandler * InteractionModelEngine::ActiveWriteHandlerAt(unsigned int aIndex) |
| { |
| unsigned int i = 0; |
| |
| for (auto & writeHandler : mWriteHandlers) |
| { |
| if (!writeHandler.IsFree()) |
| { |
| if (i == aIndex) |
| { |
| return &writeHandler; |
| } |
| i++; |
| } |
| } |
| return nullptr; |
| } |
| |
| uint32_t InteractionModelEngine::GetNumActiveWriteHandlers() const |
| { |
| uint32_t numActive = 0; |
| |
| for (auto & writeHandler : mWriteHandlers) |
| { |
| if (!writeHandler.IsFree()) |
| { |
| numActive++; |
| } |
| } |
| |
| return numActive; |
| } |
| |
| CHIP_ERROR InteractionModelEngine::ShutdownSubscription(const ScopedNodeId & aPeerNodeId, SubscriptionId aSubscriptionId) |
| { |
| for (auto * readClient = mpActiveReadClientList; readClient != nullptr; readClient = readClient->GetNextClient()) |
| { |
| if (readClient->IsSubscriptionType() && readClient->IsMatchingSubscriptionId(aSubscriptionId) && |
| readClient->GetFabricIndex() == aPeerNodeId.GetFabricIndex() && readClient->GetPeerNodeId() == aPeerNodeId.GetNodeId()) |
| { |
| readClient->Close(CHIP_NO_ERROR); |
| return CHIP_NO_ERROR; |
| } |
| } |
| |
| return CHIP_ERROR_KEY_NOT_FOUND; |
| } |
| |
| void InteractionModelEngine::ShutdownSubscriptions(FabricIndex aFabricIndex, NodeId aPeerNodeId) |
| { |
| ShutdownMatchingSubscriptions(MakeOptional(aFabricIndex), MakeOptional(aPeerNodeId)); |
| } |
| void InteractionModelEngine::ShutdownSubscriptions(FabricIndex aFabricIndex) |
| { |
| ShutdownMatchingSubscriptions(MakeOptional(aFabricIndex)); |
| } |
| |
| void InteractionModelEngine::ShutdownAllSubscriptions() |
| { |
| ShutdownMatchingSubscriptions(); |
| } |
| |
| void InteractionModelEngine::ShutdownMatchingSubscriptions(const Optional<FabricIndex> & aFabricIndex, |
| const Optional<NodeId> & aPeerNodeId) |
| { |
| // This is assuming that ReadClient::Close will not affect any other |
| // ReadClients in the list. |
| for (auto * readClient = mpActiveReadClientList; readClient != nullptr;) |
| { |
| // Grab the next client now, because we might be about to delete readClient. |
| auto * nextClient = readClient->GetNextClient(); |
| if (readClient->IsSubscriptionType()) |
| { |
| bool fabricMatches = !aFabricIndex.HasValue() || (aFabricIndex.Value() == readClient->GetFabricIndex()); |
| bool nodeIdMatches = !aPeerNodeId.HasValue() || (aPeerNodeId.Value() == readClient->GetPeerNodeId()); |
| if (fabricMatches && nodeIdMatches) |
| { |
| readClient->Close(CHIP_NO_ERROR); |
| } |
| } |
| readClient = nextClient; |
| } |
| } |
| |
| void InteractionModelEngine::OnDone(CommandHandler & apCommandObj) |
| { |
| mCommandHandlerObjs.ReleaseObject(&apCommandObj); |
| } |
| |
| void InteractionModelEngine::OnDone(ReadHandler & apReadObj) |
| { |
| // |
| // Deleting an item can shift down the contents of the underlying pool storage, |
| // rendering any tracker using positional indexes invalid. Let's reset it, |
| // based on which readHandler we are getting rid of. |
| // |
| mReportingEngine.ResetReadHandlerTracker(&apReadObj); |
| |
| mReadHandlers.ReleaseObject(&apReadObj); |
| } |
| |
| Status InteractionModelEngine::OnInvokeCommandRequest(Messaging::ExchangeContext * apExchangeContext, |
| const PayloadHeader & aPayloadHeader, System::PacketBufferHandle && aPayload, |
| bool aIsTimedInvoke) |
| { |
| CommandHandler * commandHandler = mCommandHandlerObjs.CreateObject(this); |
| if (commandHandler == nullptr) |
| { |
| ChipLogProgress(InteractionModel, "no resource for Invoke interaction"); |
| return Status::Busy; |
| } |
| commandHandler->OnInvokeCommandRequest(apExchangeContext, aPayloadHeader, std::move(aPayload), aIsTimedInvoke); |
| return Status::Success; |
| } |
| |
| CHIP_ERROR InteractionModelEngine::ParseAttributePaths(const Access::SubjectDescriptor & aSubjectDescriptor, |
| AttributePathIBs::Parser & aAttributePathListParser, |
| bool & aHasValidAttributePath, size_t & aRequestedAttributePathCount) |
| { |
| TLV::TLVReader pathReader; |
| aAttributePathListParser.GetReader(&pathReader); |
| CHIP_ERROR err = CHIP_NO_ERROR; |
| |
| aHasValidAttributePath = false; |
| aRequestedAttributePathCount = 0; |
| |
| while (CHIP_NO_ERROR == (err = pathReader.Next())) |
| { |
| VerifyOrReturnError(TLV::AnonymousTag() == pathReader.GetTag(), CHIP_ERROR_INVALID_TLV_TAG); |
| |
| AttributePathIB::Parser path; |
| // |
| // We create an iterator to point to a single item object list that tracks the path we just parsed. |
| // This avoids the 'parse all paths' approach that is employed in ReadHandler since we want to |
| // avoid allocating out of the path store during this minimal initial processing stage. |
| // |
| ObjectList<AttributePathParams> paramsList; |
| |
| ReturnErrorOnFailure(path.Init(pathReader)); |
| ReturnErrorOnFailure(path.ParsePath(paramsList.mValue)); |
| |
| if (paramsList.mValue.IsWildcardPath()) |
| { |
| AttributePathExpandIterator pathIterator(¶msList); |
| ConcreteAttributePath readPath; |
| |
| // The definition of "valid path" is "path exists and ACL allows access". The "path exists" part is handled by |
| // AttributePathExpandIterator. So we just need to check the ACL bits. |
| for (; pathIterator.Get(readPath); pathIterator.Next()) |
| { |
| Access::RequestPath requestPath{ .cluster = readPath.mClusterId, .endpoint = readPath.mEndpointId }; |
| err = Access::GetAccessControl().Check(aSubjectDescriptor, requestPath, |
| RequiredPrivilege::ForReadAttribute(readPath)); |
| if (err == CHIP_NO_ERROR) |
| { |
| aHasValidAttributePath = true; |
| break; |
| } |
| } |
| } |
| else |
| { |
| ConcreteAttributePath concretePath(paramsList.mValue.mEndpointId, paramsList.mValue.mClusterId, |
| paramsList.mValue.mAttributeId); |
| if (ConcreteAttributePathExists(concretePath)) |
| { |
| Access::RequestPath requestPath{ .cluster = concretePath.mClusterId, .endpoint = concretePath.mEndpointId }; |
| |
| err = Access::GetAccessControl().Check(aSubjectDescriptor, requestPath, |
| RequiredPrivilege::ForReadAttribute(concretePath)); |
| if (err == CHIP_NO_ERROR) |
| { |
| aHasValidAttributePath = true; |
| } |
| } |
| } |
| |
| aRequestedAttributePathCount++; |
| } |
| |
| if (err == CHIP_ERROR_END_OF_TLV) |
| { |
| err = CHIP_NO_ERROR; |
| } |
| |
| return err; |
| } |
| |
| Protocols::InteractionModel::Status InteractionModelEngine::OnReadInitialRequest(Messaging::ExchangeContext * apExchangeContext, |
| const PayloadHeader & aPayloadHeader, |
| System::PacketBufferHandle && aPayload, |
| ReadHandler::InteractionType aInteractionType) |
| { |
| ChipLogDetail(InteractionModel, "Received %s request", |
| aInteractionType == ReadHandler::InteractionType::Subscribe ? "Subscribe" : "Read"); |
| |
| // |
| // Let's first figure out if the client has sent us a subscribe request and requested we keep any existing |
| // subscriptions from that source. |
| // |
| if (aInteractionType == ReadHandler::InteractionType::Subscribe) |
| { |
| System::PacketBufferTLVReader reader; |
| bool keepExistingSubscriptions = true; |
| |
| if (apExchangeContext->GetSessionHandle()->GetFabricIndex() == kUndefinedFabricIndex) |
| { |
| // Subscriptions must be associated to a fabric. |
| return Status::UnsupportedAccess; |
| } |
| |
| reader.Init(aPayload.Retain()); |
| |
| SubscribeRequestMessage::Parser subscribeRequestParser; |
| VerifyOrReturnError(subscribeRequestParser.Init(reader) == CHIP_NO_ERROR, Status::InvalidAction); |
| |
| VerifyOrReturnError(subscribeRequestParser.GetKeepSubscriptions(&keepExistingSubscriptions) == CHIP_NO_ERROR, |
| Status::InvalidAction); |
| if (!keepExistingSubscriptions) |
| { |
| // |
| // Walk through all existing subscriptions and shut down those whose subscriber matches |
| // that which just came in. |
| // |
| mReadHandlers.ForEachActiveObject([this, apExchangeContext](ReadHandler * handler) { |
| if (handler->IsFromSubscriber(*apExchangeContext)) |
| { |
| ChipLogProgress(InteractionModel, |
| "Deleting previous subscription from NodeId: " ChipLogFormatX64 ", FabricIndex: %u", |
| ChipLogValueX64(apExchangeContext->GetSessionHandle()->AsSecureSession()->GetPeerNodeId()), |
| apExchangeContext->GetSessionHandle()->GetFabricIndex()); |
| mReadHandlers.ReleaseObject(handler); |
| } |
| |
| return Loop::Continue; |
| }); |
| } |
| |
| { |
| size_t requestedAttributePathCount = 0; |
| size_t requestedEventPathCount = 0; |
| AttributePathIBs::Parser attributePathListParser; |
| bool hasValidAttributePath = false; |
| |
| CHIP_ERROR err = subscribeRequestParser.GetAttributeRequests(&attributePathListParser); |
| if (err == CHIP_NO_ERROR) |
| { |
| auto subjectDescriptor = apExchangeContext->GetSessionHandle()->AsSecureSession()->GetSubjectDescriptor(); |
| err = ParseAttributePaths(subjectDescriptor, attributePathListParser, hasValidAttributePath, |
| requestedAttributePathCount); |
| if (err != CHIP_NO_ERROR) |
| { |
| return Status::InvalidAction; |
| } |
| } |
| else if (err != CHIP_ERROR_END_OF_TLV) |
| { |
| return Status::InvalidAction; |
| } |
| |
| EventPathIBs::Parser eventpathListParser; |
| err = subscribeRequestParser.GetEventRequests(&eventpathListParser); |
| if (err == CHIP_NO_ERROR) |
| { |
| TLV::TLVReader pathReader; |
| eventpathListParser.GetReader(&pathReader); |
| ReturnErrorCodeIf(TLV::Utilities::Count(pathReader, requestedEventPathCount, false) != CHIP_NO_ERROR, |
| Status::InvalidAction); |
| } |
| else if (err != CHIP_ERROR_END_OF_TLV) |
| { |
| return Status::InvalidAction; |
| } |
| |
| if (requestedAttributePathCount == 0 && requestedEventPathCount == 0) |
| { |
| ChipLogError(InteractionModel, |
| "Subscription from [%u:" ChipLogFormatX64 "] has no attribute or event paths. Rejecting request.", |
| apExchangeContext->GetSessionHandle()->GetFabricIndex(), |
| ChipLogValueX64(apExchangeContext->GetSessionHandle()->AsSecureSession()->GetPeerNodeId())); |
| return Status::InvalidAction; |
| } |
| |
| // |
| // TODO: We don't have an easy way to do a similar 'path expansion' for events to deduce |
| // access so for now, assume that the presence of any path in the event list means they |
| // might have some access to those events. |
| // |
| if (!hasValidAttributePath && requestedEventPathCount == 0) |
| { |
| ChipLogError(InteractionModel, |
| "Subscription from [%u:" ChipLogFormatX64 "] has no access at all. Rejecting request.", |
| apExchangeContext->GetSessionHandle()->GetFabricIndex(), |
| ChipLogValueX64(apExchangeContext->GetSessionHandle()->AsSecureSession()->GetPeerNodeId())); |
| return Status::InvalidAction; |
| } |
| |
| // The following cast is safe, since we can only hold a few tens of paths in one request. |
| if (!EnsureResourceForSubscription(apExchangeContext->GetSessionHandle()->GetFabricIndex(), requestedAttributePathCount, |
| requestedEventPathCount)) |
| { |
| return Status::PathsExhausted; |
| } |
| } |
| } |
| else |
| { |
| System::PacketBufferTLVReader reader; |
| reader.Init(aPayload.Retain()); |
| |
| ReadRequestMessage::Parser readRequestParser; |
| VerifyOrReturnError(readRequestParser.Init(reader) == CHIP_NO_ERROR, Status::InvalidAction); |
| |
| { |
| size_t requestedAttributePathCount = 0; |
| size_t requestedEventPathCount = 0; |
| AttributePathIBs::Parser attributePathListParser; |
| CHIP_ERROR err = readRequestParser.GetAttributeRequests(&attributePathListParser); |
| if (err == CHIP_NO_ERROR) |
| { |
| TLV::TLVReader pathReader; |
| attributePathListParser.GetReader(&pathReader); |
| ReturnErrorCodeIf(TLV::Utilities::Count(pathReader, requestedAttributePathCount, false) != CHIP_NO_ERROR, |
| Status::InvalidAction); |
| } |
| else if (err != CHIP_ERROR_END_OF_TLV) |
| { |
| return Status::InvalidAction; |
| } |
| EventPathIBs::Parser eventpathListParser; |
| err = readRequestParser.GetEventRequests(&eventpathListParser); |
| if (err == CHIP_NO_ERROR) |
| { |
| TLV::TLVReader pathReader; |
| eventpathListParser.GetReader(&pathReader); |
| ReturnErrorCodeIf(TLV::Utilities::Count(pathReader, requestedEventPathCount, false) != CHIP_NO_ERROR, |
| Status::InvalidAction); |
| } |
| else if (err != CHIP_ERROR_END_OF_TLV) |
| { |
| return Status::InvalidAction; |
| } |
| |
| // The following cast is safe, since we can only hold a few tens of paths in one request. |
| Status checkResult = EnsureResourceForRead(apExchangeContext->GetSessionHandle()->GetFabricIndex(), |
| requestedAttributePathCount, requestedEventPathCount); |
| if (checkResult != Status::Success) |
| { |
| return checkResult; |
| } |
| } |
| } |
| |
| // We have already reserved enough resources for read requests, and have granted enough resources for current subscriptions, so |
| // we should be able to allocate resources requested by this request. |
| ReadHandler * handler = mReadHandlers.CreateObject(*this, apExchangeContext, aInteractionType); |
| if (handler == nullptr) |
| { |
| ChipLogProgress(InteractionModel, "no resource for %s interaction", |
| aInteractionType == ReadHandler::InteractionType::Subscribe ? "Subscribe" : "Read"); |
| return Status::ResourceExhausted; |
| } |
| |
| handler->OnInitialRequest(std::move(aPayload)); |
| |
| return Status::Success; |
| } |
| |
| Protocols::InteractionModel::Status InteractionModelEngine::OnWriteRequest(Messaging::ExchangeContext * apExchangeContext, |
| const PayloadHeader & aPayloadHeader, |
| System::PacketBufferHandle && aPayload, |
| bool aIsTimedWrite) |
| { |
| ChipLogDetail(InteractionModel, "Received Write request"); |
| |
| for (auto & writeHandler : mWriteHandlers) |
| { |
| if (writeHandler.IsFree()) |
| { |
| VerifyOrReturnError(writeHandler.Init() == CHIP_NO_ERROR, Status::Busy); |
| return writeHandler.OnWriteRequest(apExchangeContext, std::move(aPayload), aIsTimedWrite); |
| } |
| } |
| ChipLogProgress(InteractionModel, "no resource for write interaction"); |
| return Status::Busy; |
| } |
| |
| CHIP_ERROR InteractionModelEngine::OnTimedRequest(Messaging::ExchangeContext * apExchangeContext, |
| const PayloadHeader & aPayloadHeader, System::PacketBufferHandle && aPayload, |
| Protocols::InteractionModel::Status & aStatus) |
| { |
| TimedHandler * handler = mTimedHandlers.CreateObject(); |
| if (handler == nullptr) |
| { |
| ChipLogProgress(InteractionModel, "no resource for Timed interaction"); |
| aStatus = Status::Busy; |
| return CHIP_ERROR_NO_MEMORY; |
| } |
| |
| // The timed handler takes over handling of this exchange and will do its |
| // own status reporting as needed. |
| aStatus = Status::Success; |
| apExchangeContext->SetDelegate(handler); |
| return handler->OnMessageReceived(apExchangeContext, aPayloadHeader, std::move(aPayload)); |
| } |
| |
| Status InteractionModelEngine::OnUnsolicitedReportData(Messaging::ExchangeContext * apExchangeContext, |
| const PayloadHeader & aPayloadHeader, System::PacketBufferHandle && aPayload) |
| { |
| System::PacketBufferTLVReader reader; |
| reader.Init(aPayload.Retain()); |
| |
| ReportDataMessage::Parser report; |
| VerifyOrReturnError(report.Init(reader) == CHIP_NO_ERROR, Status::InvalidAction); |
| |
| SubscriptionId subscriptionId = 0; |
| VerifyOrReturnError(report.GetSubscriptionId(&subscriptionId) == CHIP_NO_ERROR, Status::InvalidAction); |
| VerifyOrReturnError(report.ExitContainer() == CHIP_NO_ERROR, Status::InvalidAction); |
| |
| for (auto * readClient = mpActiveReadClientList; readClient != nullptr; readClient = readClient->GetNextClient()) |
| { |
| if (!readClient->IsSubscriptionActive()) |
| { |
| continue; |
| } |
| auto peer = apExchangeContext->GetSessionHandle()->GetPeer(); |
| if (readClient->GetFabricIndex() != peer.GetFabricIndex() || readClient->GetPeerNodeId() != peer.GetNodeId() || |
| !readClient->IsMatchingSubscriptionId(subscriptionId)) |
| { |
| continue; |
| } |
| readClient->OnUnsolicitedReportData(apExchangeContext, std::move(aPayload)); |
| return Status::Success; |
| } |
| |
| return Status::InvalidSubscription; |
| } |
| |
| CHIP_ERROR InteractionModelEngine::OnUnsolicitedMessageReceived(const PayloadHeader & payloadHeader, |
| ExchangeDelegate *& newDelegate) |
| { |
| // TODO: Implement OnUnsolicitedMessageReceived, let messaging layer dispatch message to ReadHandler/ReadClient/TimedHandler |
| // directly. |
| newDelegate = this; |
| return CHIP_NO_ERROR; |
| } |
| |
| CHIP_ERROR InteractionModelEngine::OnMessageReceived(Messaging::ExchangeContext * apExchangeContext, |
| const PayloadHeader & aPayloadHeader, System::PacketBufferHandle && aPayload) |
| { |
| using namespace Protocols::InteractionModel; |
| |
| Protocols::InteractionModel::Status status = Status::Failure; |
| |
| // Group Message can only be an InvokeCommandRequest or WriteRequest |
| if (apExchangeContext->IsGroupExchangeContext() && |
| !aPayloadHeader.HasMessageType(Protocols::InteractionModel::MsgType::InvokeCommandRequest) && |
| !aPayloadHeader.HasMessageType(Protocols::InteractionModel::MsgType::WriteRequest)) |
| { |
| ChipLogProgress(InteractionModel, "Msg type %d not supported for group message", aPayloadHeader.GetMessageType()); |
| return CHIP_NO_ERROR; |
| } |
| |
| if (aPayloadHeader.HasMessageType(Protocols::InteractionModel::MsgType::InvokeCommandRequest)) |
| { |
| status = OnInvokeCommandRequest(apExchangeContext, aPayloadHeader, std::move(aPayload), /* aIsTimedInvoke = */ false); |
| } |
| else if (aPayloadHeader.HasMessageType(Protocols::InteractionModel::MsgType::ReadRequest)) |
| { |
| status = OnReadInitialRequest(apExchangeContext, aPayloadHeader, std::move(aPayload), ReadHandler::InteractionType::Read); |
| } |
| else if (aPayloadHeader.HasMessageType(Protocols::InteractionModel::MsgType::WriteRequest)) |
| { |
| status = OnWriteRequest(apExchangeContext, aPayloadHeader, std::move(aPayload), /* aIsTimedWrite = */ false); |
| } |
| else if (aPayloadHeader.HasMessageType(Protocols::InteractionModel::MsgType::SubscribeRequest)) |
| { |
| status = |
| OnReadInitialRequest(apExchangeContext, aPayloadHeader, std::move(aPayload), ReadHandler::InteractionType::Subscribe); |
| } |
| else if (aPayloadHeader.HasMessageType(Protocols::InteractionModel::MsgType::ReportData)) |
| { |
| status = OnUnsolicitedReportData(apExchangeContext, aPayloadHeader, std::move(aPayload)); |
| } |
| else if (aPayloadHeader.HasMessageType(MsgType::TimedRequest)) |
| { |
| OnTimedRequest(apExchangeContext, aPayloadHeader, std::move(aPayload), status); |
| } |
| else |
| { |
| ChipLogProgress(InteractionModel, "Msg type %d not supported", aPayloadHeader.GetMessageType()); |
| status = Status::InvalidAction; |
| } |
| |
| if (status != Status::Success && !apExchangeContext->IsGroupExchangeContext()) |
| { |
| return StatusResponse::Send(status, apExchangeContext, false /*aExpectResponse*/); |
| } |
| |
| return CHIP_NO_ERROR; |
| } |
| |
| void InteractionModelEngine::OnResponseTimeout(Messaging::ExchangeContext * ec) |
| { |
| ChipLogError(InteractionModel, "Time out! Failed to receive IM response from Exchange: " ChipLogFormatExchange, |
| ChipLogValueExchange(ec)); |
| } |
| |
| void InteractionModelEngine::AddReadClient(ReadClient * apReadClient) |
| { |
| apReadClient->SetNextClient(mpActiveReadClientList); |
| mpActiveReadClientList = apReadClient; |
| } |
| |
| bool InteractionModelEngine::TrimFabricForSubscriptions(FabricIndex aFabricIndex, bool aForceEvict) |
| { |
| const size_t pathPoolCapacity = GetPathPoolCapacityForSubscriptions(); |
| const size_t readHandlerPoolCapacity = GetReadHandlerPoolCapacityForSubscriptions(); |
| |
| uint8_t fabricCount = mpFabricTable->FabricCount(); |
| size_t attributePathsSubscribedByCurrentFabric = 0; |
| size_t eventPathsSubscribedByCurrentFabric = 0; |
| size_t subscriptionsEstablishedByCurrentFabric = 0; |
| |
| if (fabricCount == 0) |
| { |
| return false; |
| } |
| |
| // Note: This is OK only when we have assumed the fabricCount is not zero. Should be revised when adding support to |
| // subscriptions on PASE sessions. |
| size_t perFabricPathCapacity = pathPoolCapacity / static_cast<size_t>(fabricCount); |
| size_t perFabricSubscriptionCapacity = readHandlerPoolCapacity / static_cast<size_t>(fabricCount); |
| |
| ReadHandler * candidate = nullptr; |
| size_t candidateAttributePathsUsed = 0; |
| size_t candidateEventPathsUsed = 0; |
| |
| // It is safe to use & here since this function will be called on current stack. |
| mReadHandlers.ForEachActiveObject([&](ReadHandler * handler) { |
| if (handler->GetAccessingFabricIndex() != aFabricIndex || !handler->IsType(ReadHandler::InteractionType::Subscribe)) |
| { |
| return Loop::Continue; |
| } |
| |
| size_t attributePathsUsed = handler->GetAttributePathCount(); |
| size_t eventPathsUsed = handler->GetEventPathCount(); |
| |
| attributePathsSubscribedByCurrentFabric += attributePathsUsed; |
| eventPathsSubscribedByCurrentFabric += eventPathsUsed; |
| subscriptionsEstablishedByCurrentFabric++; |
| |
| if (candidate == nullptr) |
| { |
| candidate = handler; |
| } |
| // This handler uses more resources than the one we picked before. |
| else if ((attributePathsUsed > perFabricPathCapacity || eventPathsUsed > perFabricPathCapacity) && |
| (candidateAttributePathsUsed <= perFabricPathCapacity && candidateEventPathsUsed <= perFabricPathCapacity)) |
| { |
| candidate = handler; |
| candidateAttributePathsUsed = attributePathsUsed; |
| candidateEventPathsUsed = eventPathsUsed; |
| } |
| // This handler is older than the one we picked before. |
| else if (handler->GetTransactionStartGeneration() < candidate->GetTransactionStartGeneration() && |
| // And the level of resource usage is the same (both exceed or neither exceed) |
| ((attributePathsUsed > perFabricPathCapacity || eventPathsUsed > perFabricPathCapacity) == |
| (candidateAttributePathsUsed > perFabricPathCapacity || candidateEventPathsUsed > perFabricPathCapacity))) |
| { |
| candidate = handler; |
| } |
| return Loop::Continue; |
| }); |
| |
| if (candidate != nullptr && |
| (aForceEvict || attributePathsSubscribedByCurrentFabric > perFabricPathCapacity || |
| eventPathsSubscribedByCurrentFabric > perFabricPathCapacity || |
| subscriptionsEstablishedByCurrentFabric > perFabricSubscriptionCapacity)) |
| { |
| SubscriptionId subId; |
| candidate->GetSubscriptionId(subId); |
| ChipLogProgress(DataManagement, "Evicting Subscription ID %u:0x%" PRIx32, candidate->GetSubjectDescriptor().fabricIndex, |
| subId); |
| candidate->Close(); |
| return true; |
| } |
| return false; |
| } |
| |
| bool InteractionModelEngine::EnsureResourceForSubscription(FabricIndex aFabricIndex, size_t aRequestedAttributePathCount, |
| size_t aRequestedEventPathCount) |
| { |
| #if CHIP_SYSTEM_CONFIG_POOL_USE_HEAP && !CHIP_CONFIG_IM_FORCE_FABRIC_QUOTA_CHECK |
| #if CONFIG_BUILD_FOR_HOST_UNIT_TEST |
| const bool allowUnlimited = !mForceHandlerQuota; |
| #else // CONFIG_BUILD_FOR_HOST_UNIT_TEST |
| // If the resources are allocated on the heap, we should be able to handle as many Read / Subscribe requests as possible. |
| const bool allowUnlimited = true; |
| #endif // CONFIG_BUILD_FOR_HOST_UNIT_TEST |
| #else // CHIP_SYSTEM_CONFIG_POOL_USE_HEAP && !CHIP_CONFIG_IM_FORCE_FABRIC_QUOTA_CHECK |
| const bool allowUnlimited = false; |
| #endif // CHIP_SYSTEM_CONFIG_POOL_USE_HEAP && !CHIP_CONFIG_IM_FORCE_FABRIC_QUOTA_CHECK |
| |
| // Don't couple with read requests, always reserve enough resource for read requests. |
| |
| const size_t pathPoolCapacity = GetPathPoolCapacityForSubscriptions(); |
| const size_t readHandlerPoolCapacity = GetReadHandlerPoolCapacityForSubscriptions(); |
| |
| // If we return early here, the compiler will complain about the unreachable code, so we add a always-true check. |
| const size_t attributePathCap = allowUnlimited ? SIZE_MAX : pathPoolCapacity; |
| const size_t eventPathCap = allowUnlimited ? SIZE_MAX : pathPoolCapacity; |
| const size_t readHandlerCap = allowUnlimited ? SIZE_MAX : readHandlerPoolCapacity; |
| |
| size_t usedAttributePaths = 0; |
| size_t usedEventPaths = 0; |
| size_t usedReadHandlers = 0; |
| |
| auto countResourceUsage = [&]() { |
| usedAttributePaths = 0; |
| usedEventPaths = 0; |
| usedReadHandlers = 0; |
| mReadHandlers.ForEachActiveObject([&](auto * handler) { |
| if (!handler->IsType(ReadHandler::InteractionType::Subscribe)) |
| { |
| return Loop::Continue; |
| } |
| usedAttributePaths += handler->GetAttributePathCount(); |
| usedEventPaths += handler->GetEventPathCount(); |
| usedReadHandlers++; |
| return Loop::Continue; |
| }); |
| }; |
| |
| countResourceUsage(); |
| |
| if (usedAttributePaths + aRequestedAttributePathCount <= attributePathCap && |
| usedEventPaths + aRequestedEventPathCount <= eventPathCap && usedReadHandlers < readHandlerCap) |
| { |
| // We have enough resources, then we serve the requests in a best-effort manner. |
| return true; |
| } |
| |
| if ((aRequestedAttributePathCount > kMinSupportedPathsPerSubscription && |
| usedAttributePaths + aRequestedAttributePathCount > attributePathCap) || |
| (aRequestedEventPathCount > kMinSupportedPathsPerSubscription && usedEventPaths + aRequestedEventPathCount > eventPathCap)) |
| { |
| // We cannot offer enough resources, and the subscription is requesting more than the spec limit. |
| return false; |
| } |
| |
| const auto evictAndUpdateResourceUsage = [&](FabricIndex fabricIndex, bool forceEvict) { |
| bool ret = TrimFabricForSubscriptions(fabricIndex, forceEvict); |
| countResourceUsage(); |
| return ret; |
| }; |
| |
| // |
| // At this point, we have an inbound request that respects minimas but we still don't have enough resources to handle it. Which |
| // means that we definitely have handlers on existing fabrics that are over limits and need to evict at least one of them to |
| // make space. |
| // |
| // There might be cases that one fabric has lots of subscriptions with one interested path, while the other fabrics are not |
| // using excess resources. So we need to do this multiple times until we have enough room or no fabrics are using excess |
| // resources. |
| // |
| bool didEvictHandler = true; |
| while (didEvictHandler) |
| { |
| didEvictHandler = false; |
| for (const auto & fabric : *mpFabricTable) |
| { |
| // The resources are enough to serve this request, do not evict anything. |
| if (usedAttributePaths + aRequestedAttributePathCount <= attributePathCap && |
| usedEventPaths + aRequestedEventPathCount <= eventPathCap && usedReadHandlers < readHandlerCap) |
| { |
| break; |
| } |
| didEvictHandler = didEvictHandler || evictAndUpdateResourceUsage(fabric.GetFabricIndex(), false); |
| } |
| } |
| |
| // The above loop cannot guarantee the resources for the new subscriptions when the resource usage from all fabrics are exactly |
| // within the quota (which means we have exactly used all resources). Evict (from the large subscriptions first then from |
| // oldest) subscriptions from the current fabric until we have enough resource for the new subscription. |
| didEvictHandler = true; |
| while ((usedAttributePaths + aRequestedAttributePathCount > attributePathCap || |
| usedEventPaths + aRequestedEventPathCount > eventPathCap || usedReadHandlers >= readHandlerCap) && |
| // Avoid infinity loop |
| didEvictHandler) |
| { |
| didEvictHandler = evictAndUpdateResourceUsage(aFabricIndex, true); |
| } |
| |
| // If didEvictHandler is false, means the loop above evicted all subscriptions from the current fabric but we still don't have |
| // enough resources for the new subscription, this should never happen. |
| // This is safe as long as we have rejected subscriptions without a fabric associated (with a PASE session) before. |
| // Note: Spec#5141: should reject subscription requests on PASE sessions. |
| VerifyOrDieWithMsg(didEvictHandler, DataManagement, "Failed to get required resources by evicting existing subscriptions."); |
| |
| // We have ensured enough resources by the logic above. |
| return true; |
| } |
| |
| bool InteractionModelEngine::TrimFabricForRead(FabricIndex aFabricIndex) |
| { |
| const size_t guaranteedReadRequestsPerFabric = GetGuaranteedReadRequestsPerFabric(); |
| const size_t minSupportedPathsPerFabricForRead = guaranteedReadRequestsPerFabric * kMinSupportedPathsPerReadRequest; |
| |
| size_t attributePathsUsedByCurrentFabric = 0; |
| size_t eventPathsUsedByCurrentFabric = 0; |
| size_t readTransactionsOnCurrentFabric = 0; |
| |
| ReadHandler * candidate = nullptr; |
| size_t candidateAttributePathsUsed = 0; |
| size_t candidateEventPathsUsed = 0; |
| |
| // It is safe to use & here since this function will be called on current stack. |
| mReadHandlers.ForEachActiveObject([&](ReadHandler * handler) { |
| if (handler->GetAccessingFabricIndex() != aFabricIndex || !handler->IsType(ReadHandler::InteractionType::Read)) |
| { |
| return Loop::Continue; |
| } |
| |
| size_t attributePathsUsed = handler->GetAttributePathCount(); |
| size_t eventPathsUsed = handler->GetEventPathCount(); |
| |
| attributePathsUsedByCurrentFabric += attributePathsUsed; |
| eventPathsUsedByCurrentFabric += eventPathsUsed; |
| readTransactionsOnCurrentFabric++; |
| |
| if (candidate == nullptr) |
| { |
| candidate = handler; |
| } |
| // Oversized read handlers will be evicted first. |
| else if ((attributePathsUsed > kMinSupportedPathsPerReadRequest || eventPathsUsed > kMinSupportedPathsPerReadRequest) && |
| (candidateAttributePathsUsed <= kMinSupportedPathsPerReadRequest && |
| candidateEventPathsUsed <= kMinSupportedPathsPerReadRequest)) |
| { |
| candidate = handler; |
| } |
| // Read Handlers are "first come first served", so we give eariler read transactions a higher priority. |
| else if (handler->GetTransactionStartGeneration() > candidate->GetTransactionStartGeneration() && |
| // And the level of resource usage is the same (both exceed or neither exceed) |
| ((attributePathsUsed > kMinSupportedPathsPerReadRequest || eventPathsUsed > kMinSupportedPathsPerReadRequest) == |
| (candidateAttributePathsUsed > kMinSupportedPathsPerReadRequest || |
| candidateEventPathsUsed > kMinSupportedPathsPerReadRequest))) |
| { |
| candidate = handler; |
| } |
| |
| if (candidate == handler) |
| { |
| candidateAttributePathsUsed = attributePathsUsed; |
| candidateEventPathsUsed = eventPathsUsed; |
| } |
| return Loop::Continue; |
| }); |
| |
| if (candidate != nullptr && |
| ((attributePathsUsedByCurrentFabric > minSupportedPathsPerFabricForRead || |
| eventPathsUsedByCurrentFabric > minSupportedPathsPerFabricForRead || |
| readTransactionsOnCurrentFabric > guaranteedReadRequestsPerFabric) || |
| // Always evict the transactions on PASE sessions if the fabric table is full. |
| (aFabricIndex == kUndefinedFabricIndex && mpFabricTable->FabricCount() == GetConfigMaxFabrics()))) |
| { |
| candidate->Close(); |
| return true; |
| } |
| return false; |
| } |
| |
| Protocols::InteractionModel::Status InteractionModelEngine::EnsureResourceForRead(FabricIndex aFabricIndex, |
| size_t aRequestedAttributePathCount, |
| size_t aRequestedEventPathCount) |
| { |
| #if CHIP_SYSTEM_CONFIG_POOL_USE_HEAP && !CHIP_CONFIG_IM_FORCE_FABRIC_QUOTA_CHECK |
| #if CONFIG_BUILD_FOR_HOST_UNIT_TEST |
| const bool allowUnlimited = !mForceHandlerQuota; |
| #else // CONFIG_BUILD_FOR_HOST_UNIT_TEST |
| // If the resources are allocated on the heap, we should be able to handle as many Read / Subscribe requests as possible. |
| const bool allowUnlimited = true; |
| #endif // CONFIG_BUILD_FOR_HOST_UNIT_TEST |
| #else // CHIP_SYSTEM_CONFIG_POOL_USE_HEAP && !CHIP_CONFIG_IM_FORCE_FABRIC_QUOTA_CHECK |
| const bool allowUnlimited = false; |
| #endif // CHIP_SYSTEM_CONFIG_POOL_USE_HEAP && !CHIP_CONFIG_IM_FORCE_FABRIC_QUOTA_CHECK |
| |
| // If we return early here, the compiler will complain about the unreachable code, so we add a always-true check. |
| const size_t attributePathCap = allowUnlimited ? SIZE_MAX : GetPathPoolCapacityForReads(); |
| const size_t eventPathCap = allowUnlimited ? SIZE_MAX : GetPathPoolCapacityForReads(); |
| const size_t readHandlerCap = allowUnlimited ? SIZE_MAX : GetReadHandlerPoolCapacityForReads(); |
| |
| const size_t guaranteedReadRequestsPerFabric = GetGuaranteedReadRequestsPerFabric(); |
| const size_t guaranteedPathsPerFabric = kMinSupportedPathsPerReadRequest * guaranteedReadRequestsPerFabric; |
| |
| size_t usedAttributePaths = 0; |
| size_t usedEventPaths = 0; |
| size_t usedReadHandlers = 0; |
| |
| auto countResourceUsage = [&]() { |
| usedAttributePaths = 0; |
| usedEventPaths = 0; |
| usedReadHandlers = 0; |
| mReadHandlers.ForEachActiveObject([&](auto * handler) { |
| if (!handler->IsType(ReadHandler::InteractionType::Read)) |
| { |
| return Loop::Continue; |
| } |
| usedAttributePaths += handler->GetAttributePathCount(); |
| usedEventPaths += handler->GetEventPathCount(); |
| usedReadHandlers++; |
| return Loop::Continue; |
| }); |
| }; |
| |
| auto haveEnoughResourcesForTheRequest = [&]() { |
| return usedAttributePaths + aRequestedAttributePathCount <= attributePathCap && |
| usedEventPaths + aRequestedEventPathCount <= eventPathCap && usedReadHandlers < readHandlerCap; |
| }; |
| |
| countResourceUsage(); |
| |
| if (haveEnoughResourcesForTheRequest()) |
| { |
| // We have enough resources, then we serve the requests in a best-effort manner. |
| return Status::Success; |
| } |
| |
| if ((aRequestedAttributePathCount > kMinSupportedPathsPerReadRequest && |
| usedAttributePaths + aRequestedAttributePathCount > attributePathCap) || |
| (aRequestedEventPathCount > kMinSupportedPathsPerReadRequest && usedEventPaths + aRequestedEventPathCount > eventPathCap)) |
| { |
| // We cannot offer enough resources, and the read transaction is requesting more than the spec limit. |
| return Status::PathsExhausted; |
| } |
| |
| // If we have commissioned CHIP_CONFIG_MAX_FABRICS already, and this transaction doesn't have an associated fabric index, reject |
| // the request if we don't have sufficient resources for this request. |
| if (mpFabricTable->FabricCount() == GetConfigMaxFabrics() && aFabricIndex == kUndefinedFabricIndex) |
| { |
| return Status::Busy; |
| } |
| |
| size_t usedAttributePathsInFabric = 0; |
| size_t usedEventPathsInFabric = 0; |
| size_t usedReadHandlersInFabric = 0; |
| mReadHandlers.ForEachActiveObject([&](auto * handler) { |
| if (!handler->IsType(ReadHandler::InteractionType::Read) || handler->GetAccessingFabricIndex() != aFabricIndex) |
| { |
| return Loop::Continue; |
| } |
| usedAttributePathsInFabric += handler->GetAttributePathCount(); |
| usedEventPathsInFabric += handler->GetEventPathCount(); |
| usedReadHandlersInFabric++; |
| return Loop::Continue; |
| }); |
| |
| // Busy, since there are already some read requests ongoing on this fabric, please retry later. |
| if (usedAttributePathsInFabric + aRequestedAttributePathCount > guaranteedPathsPerFabric || |
| usedEventPathsInFabric + aRequestedEventPathCount > guaranteedPathsPerFabric || |
| usedReadHandlersInFabric >= guaranteedReadRequestsPerFabric) |
| { |
| return Status::Busy; |
| } |
| |
| const auto evictAndUpdateResourceUsage = [&](FabricIndex fabricIndex) { |
| bool ret = TrimFabricForRead(fabricIndex); |
| countResourceUsage(); |
| return ret; |
| }; |
| |
| // |
| // At this point, we have an inbound request that respects minimas but we still don't have enough resources to handle it. Which |
| // means that we definitely have handlers on existing fabrics that are over limits and need to evict at least one of them to |
| // make space. |
| // |
| bool didEvictHandler = true; |
| while (didEvictHandler) |
| { |
| didEvictHandler = false; |
| didEvictHandler = didEvictHandler || evictAndUpdateResourceUsage(kUndefinedFabricIndex); |
| if (haveEnoughResourcesForTheRequest()) |
| { |
| break; |
| } |
| // If the fabric table is full, we won't evict read requests from normal fabrics before we have evicted all read requests |
| // from PASE sessions. |
| if (mpFabricTable->FabricCount() == GetConfigMaxFabrics() && didEvictHandler) |
| { |
| continue; |
| } |
| for (const auto & fabric : *mpFabricTable) |
| { |
| didEvictHandler = didEvictHandler || evictAndUpdateResourceUsage(fabric.GetFabricIndex()); |
| // If we now have enough resources to serve this request, stop evicting things. |
| if (haveEnoughResourcesForTheRequest()) |
| { |
| break; |
| } |
| } |
| } |
| |
| // Now all fabrics are not oversized (since we have trimmed the oversized fabrics in the loop above), and the read handler is |
| // also not oversized, we should be able to handle this read transaction. |
| VerifyOrDie(haveEnoughResourcesForTheRequest()); |
| |
| return Status::Success; |
| } |
| |
| void InteractionModelEngine::RemoveReadClient(ReadClient * apReadClient) |
| { |
| ReadClient * pPrevListItem = nullptr; |
| ReadClient * pCurListItem = mpActiveReadClientList; |
| |
| while (pCurListItem != apReadClient) |
| { |
| // |
| // Item must exist in this tracker list. If not, there's a bug somewhere. |
| // |
| VerifyOrDie(pCurListItem != nullptr); |
| |
| pPrevListItem = pCurListItem; |
| pCurListItem = pCurListItem->GetNextClient(); |
| } |
| |
| if (pPrevListItem) |
| { |
| pPrevListItem->SetNextClient(apReadClient->GetNextClient()); |
| } |
| else |
| { |
| mpActiveReadClientList = apReadClient->GetNextClient(); |
| } |
| |
| apReadClient->SetNextClient(nullptr); |
| } |
| |
| size_t InteractionModelEngine::GetNumActiveReadClients() |
| { |
| ReadClient * pListItem = mpActiveReadClientList; |
| size_t count = 0; |
| |
| while (pListItem) |
| { |
| pListItem = pListItem->GetNextClient(); |
| count++; |
| } |
| |
| return count; |
| } |
| |
| bool InteractionModelEngine::InActiveReadClientList(ReadClient * apReadClient) |
| { |
| ReadClient * pListItem = mpActiveReadClientList; |
| |
| while (pListItem) |
| { |
| if (pListItem == apReadClient) |
| { |
| return true; |
| } |
| |
| pListItem = pListItem->GetNextClient(); |
| } |
| |
| return false; |
| } |
| |
| bool InteractionModelEngine::HasConflictWriteRequests(const WriteHandler * apWriteHandler, const ConcreteAttributePath & aPath) |
| { |
| for (auto & writeHandler : mWriteHandlers) |
| { |
| if (writeHandler.IsFree() || &writeHandler == apWriteHandler) |
| { |
| continue; |
| } |
| if (writeHandler.IsCurrentlyProcessingWritePath(aPath)) |
| { |
| return true; |
| } |
| } |
| return false; |
| } |
| |
| void InteractionModelEngine::ReleaseAttributePathList(ObjectList<AttributePathParams> *& aAttributePathList) |
| { |
| ReleasePool(aAttributePathList, mAttributePathPool); |
| } |
| |
| CHIP_ERROR InteractionModelEngine::PushFrontAttributePathList(ObjectList<AttributePathParams> *& aAttributePathList, |
| AttributePathParams & aAttributePath) |
| { |
| CHIP_ERROR err = PushFront(aAttributePathList, aAttributePath, mAttributePathPool); |
| if (err == CHIP_ERROR_NO_MEMORY) |
| { |
| ChipLogError(InteractionModel, "AttributePath pool full"); |
| return CHIP_IM_GLOBAL_STATUS(PathsExhausted); |
| } |
| return err; |
| } |
| |
| void InteractionModelEngine::RemoveDuplicateConcreteAttributePath(ObjectList<AttributePathParams> *& aAttributePaths) |
| { |
| ObjectList<AttributePathParams> * prev = nullptr; |
| auto * path1 = aAttributePaths; |
| |
| while (path1 != nullptr) |
| { |
| bool duplicate = false; |
| // skip all wildcard paths and invalid concrete attribute |
| if (path1->mValue.IsWildcardPath() || |
| !emberAfContainsAttribute(path1->mValue.mEndpointId, path1->mValue.mClusterId, path1->mValue.mAttributeId)) |
| { |
| prev = path1; |
| path1 = path1->mpNext; |
| continue; |
| } |
| |
| // Check whether a wildcard path expands to something that includes this concrete path. |
| for (auto * path2 = aAttributePaths; path2 != nullptr; path2 = path2->mpNext) |
| { |
| if (path2 == path1) |
| { |
| continue; |
| } |
| |
| if (path2->mValue.IsWildcardPath() && path2->mValue.IsAttributePathSupersetOf(path1->mValue)) |
| { |
| duplicate = true; |
| break; |
| } |
| } |
| |
| // if path1 duplicates something from wildcard expansion, discard path1 |
| if (!duplicate) |
| { |
| prev = path1; |
| path1 = path1->mpNext; |
| continue; |
| } |
| |
| if (path1 == aAttributePaths) |
| { |
| aAttributePaths = path1->mpNext; |
| mAttributePathPool.ReleaseObject(path1); |
| path1 = aAttributePaths; |
| } |
| else |
| { |
| prev->mpNext = path1->mpNext; |
| mAttributePathPool.ReleaseObject(path1); |
| path1 = prev->mpNext; |
| } |
| } |
| } |
| |
| void InteractionModelEngine::ReleaseEventPathList(ObjectList<EventPathParams> *& aEventPathList) |
| { |
| ReleasePool(aEventPathList, mEventPathPool); |
| } |
| |
| CHIP_ERROR InteractionModelEngine::PushFrontEventPathParamsList(ObjectList<EventPathParams> *& aEventPathList, |
| EventPathParams & aEventPath) |
| { |
| CHIP_ERROR err = PushFront(aEventPathList, aEventPath, mEventPathPool); |
| if (err == CHIP_ERROR_NO_MEMORY) |
| { |
| ChipLogError(InteractionModel, "EventPath pool full"); |
| return CHIP_IM_GLOBAL_STATUS(PathsExhausted); |
| } |
| return err; |
| } |
| |
| void InteractionModelEngine::ReleaseDataVersionFilterList(ObjectList<DataVersionFilter> *& aDataVersionFilterList) |
| { |
| ReleasePool(aDataVersionFilterList, mDataVersionFilterPool); |
| } |
| |
| CHIP_ERROR InteractionModelEngine::PushFrontDataVersionFilterList(ObjectList<DataVersionFilter> *& aDataVersionFilterList, |
| DataVersionFilter & aDataVersionFilter) |
| { |
| CHIP_ERROR err = PushFront(aDataVersionFilterList, aDataVersionFilter, mDataVersionFilterPool); |
| if (err == CHIP_ERROR_NO_MEMORY) |
| { |
| ChipLogError(InteractionModel, "DataVersionFilter pool full, ignore this filter"); |
| err = CHIP_NO_ERROR; |
| } |
| return err; |
| } |
| |
| template <typename T, size_t N> |
| void InteractionModelEngine::ReleasePool(ObjectList<T> *& aObjectList, ObjectPool<ObjectList<T>, N> & aObjectPool) |
| { |
| ObjectList<T> * current = aObjectList; |
| while (current != nullptr) |
| { |
| ObjectList<T> * nextObject = current->mpNext; |
| aObjectPool.ReleaseObject(current); |
| current = nextObject; |
| } |
| |
| aObjectList = nullptr; |
| } |
| |
| template <typename T, size_t N> |
| CHIP_ERROR InteractionModelEngine::PushFront(ObjectList<T> *& aObjectList, T & aData, ObjectPool<ObjectList<T>, N> & aObjectPool) |
| { |
| ObjectList<T> * object = aObjectPool.CreateObject(); |
| if (object == nullptr) |
| { |
| return CHIP_ERROR_NO_MEMORY; |
| } |
| object->mValue = aData; |
| object->mpNext = aObjectList; |
| aObjectList = object; |
| return CHIP_NO_ERROR; |
| } |
| |
| void InteractionModelEngine::DispatchCommand(CommandHandler & apCommandObj, const ConcreteCommandPath & aCommandPath, |
| TLV::TLVReader & apPayload) |
| { |
| CommandHandlerInterface * handler = FindCommandHandler(aCommandPath.mEndpointId, aCommandPath.mClusterId); |
| |
| if (handler) |
| { |
| CommandHandlerInterface::HandlerContext context(apCommandObj, aCommandPath, apPayload); |
| handler->InvokeCommand(context); |
| |
| // |
| // If the command was handled, don't proceed any further and return successfully. |
| // |
| if (context.mCommandHandled) |
| { |
| return; |
| } |
| } |
| |
| DispatchSingleClusterCommand(aCommandPath, apPayload, &apCommandObj); |
| } |
| |
| Protocols::InteractionModel::Status InteractionModelEngine::CommandExists(const ConcreteCommandPath & aCommandPath) |
| { |
| return ServerClusterCommandExists(aCommandPath); |
| } |
| |
| CHIP_ERROR InteractionModelEngine::RegisterCommandHandler(CommandHandlerInterface * handler) |
| { |
| VerifyOrReturnError(handler != nullptr, CHIP_ERROR_INVALID_ARGUMENT); |
| |
| for (auto * cur = mCommandHandlerList; cur; cur = cur->GetNext()) |
| { |
| if (cur->Matches(*handler)) |
| { |
| ChipLogError(InteractionModel, "Duplicate command handler registration failed"); |
| return CHIP_ERROR_INCORRECT_STATE; |
| } |
| } |
| |
| handler->SetNext(mCommandHandlerList); |
| mCommandHandlerList = handler; |
| |
| return CHIP_NO_ERROR; |
| } |
| |
| void InteractionModelEngine::UnregisterCommandHandlers(EndpointId endpointId) |
| { |
| CommandHandlerInterface * prev = nullptr; |
| |
| for (auto * cur = mCommandHandlerList; cur; cur = cur->GetNext()) |
| { |
| if (cur->MatchesEndpoint(endpointId)) |
| { |
| if (prev == nullptr) |
| { |
| mCommandHandlerList = cur->GetNext(); |
| } |
| else |
| { |
| prev->SetNext(cur->GetNext()); |
| } |
| |
| cur->SetNext(nullptr); |
| } |
| else |
| { |
| prev = cur; |
| } |
| } |
| } |
| |
| CHIP_ERROR InteractionModelEngine::UnregisterCommandHandler(CommandHandlerInterface * handler) |
| { |
| VerifyOrReturnError(handler != nullptr, CHIP_ERROR_INVALID_ARGUMENT); |
| CommandHandlerInterface * prev = nullptr; |
| |
| for (auto * cur = mCommandHandlerList; cur; cur = cur->GetNext()) |
| { |
| if (cur->Matches(*handler)) |
| { |
| if (prev == nullptr) |
| { |
| mCommandHandlerList = cur->GetNext(); |
| } |
| else |
| { |
| prev->SetNext(cur->GetNext()); |
| } |
| |
| cur->SetNext(nullptr); |
| |
| return CHIP_NO_ERROR; |
| } |
| |
| prev = cur; |
| } |
| |
| return CHIP_ERROR_KEY_NOT_FOUND; |
| } |
| |
| CommandHandlerInterface * InteractionModelEngine::FindCommandHandler(EndpointId endpointId, ClusterId clusterId) |
| { |
| for (auto * cur = mCommandHandlerList; cur; cur = cur->GetNext()) |
| { |
| if (cur->Matches(endpointId, clusterId)) |
| { |
| return cur; |
| } |
| } |
| |
| return nullptr; |
| } |
| |
| void InteractionModelEngine::OnTimedInteractionFailed(TimedHandler * apTimedHandler) |
| { |
| mTimedHandlers.ReleaseObject(apTimedHandler); |
| } |
| |
| void InteractionModelEngine::OnTimedInvoke(TimedHandler * apTimedHandler, Messaging::ExchangeContext * apExchangeContext, |
| const PayloadHeader & aPayloadHeader, System::PacketBufferHandle && aPayload) |
| { |
| using namespace Protocols::InteractionModel; |
| |
| // Reset the ourselves as the exchange delegate for now, to match what we'd |
| // do with an initial unsolicited invoke. |
| apExchangeContext->SetDelegate(this); |
| mTimedHandlers.ReleaseObject(apTimedHandler); |
| |
| VerifyOrDie(aPayloadHeader.HasMessageType(MsgType::InvokeCommandRequest)); |
| VerifyOrDie(!apExchangeContext->IsGroupExchangeContext()); |
| |
| Status status = OnInvokeCommandRequest(apExchangeContext, aPayloadHeader, std::move(aPayload), /* aIsTimedInvoke = */ true); |
| if (status != Status::Success) |
| { |
| StatusResponse::Send(status, apExchangeContext, /* aExpectResponse = */ false); |
| } |
| } |
| |
| void InteractionModelEngine::OnTimedWrite(TimedHandler * apTimedHandler, Messaging::ExchangeContext * apExchangeContext, |
| const PayloadHeader & aPayloadHeader, System::PacketBufferHandle && aPayload) |
| { |
| using namespace Protocols::InteractionModel; |
| |
| // Reset the ourselves as the exchange delegate for now, to match what we'd |
| // do with an initial unsolicited write. |
| apExchangeContext->SetDelegate(this); |
| mTimedHandlers.ReleaseObject(apTimedHandler); |
| |
| VerifyOrDie(aPayloadHeader.HasMessageType(MsgType::WriteRequest)); |
| VerifyOrDie(!apExchangeContext->IsGroupExchangeContext()); |
| |
| Status status = OnWriteRequest(apExchangeContext, aPayloadHeader, std::move(aPayload), /* aIsTimedWrite = */ true); |
| if (status != Status::Success) |
| { |
| StatusResponse::Send(status, apExchangeContext, /* aExpectResponse = */ false); |
| } |
| } |
| |
| bool InteractionModelEngine::HasActiveRead() |
| { |
| return ((mReadHandlers.ForEachActiveObject([](ReadHandler * handler) { |
| if (handler->IsType(ReadHandler::InteractionType::Read)) |
| { |
| return Loop::Break; |
| } |
| |
| return Loop::Continue; |
| }) == Loop::Break)); |
| } |
| |
| uint16_t InteractionModelEngine::GetMinGuaranteedSubscriptionsPerFabric() const |
| { |
| #if CHIP_SYSTEM_CONFIG_POOL_USE_HEAP |
| return UINT16_MAX; |
| #else |
| return static_cast<uint16_t>( |
| min(GetReadHandlerPoolCapacityForSubscriptions() / GetConfigMaxFabrics(), static_cast<size_t>(UINT16_MAX))); |
| #endif |
| } |
| |
| size_t InteractionModelEngine::GetNumDirtySubscriptions() const |
| { |
| size_t numDirtySubscriptions = 0; |
| mReadHandlers.ForEachActiveObject([&](const auto readHandler) { |
| if (readHandler->IsType(ReadHandler::InteractionType::Subscribe) && readHandler->IsDirty()) |
| { |
| numDirtySubscriptions++; |
| } |
| return Loop::Continue; |
| }); |
| return numDirtySubscriptions; |
| } |
| |
| void InteractionModelEngine::OnFabricRemoved(const FabricTable & fabricTable, FabricIndex fabricIndex) |
| { |
| mReadHandlers.ForEachActiveObject([fabricIndex](ReadHandler * handler) { |
| if (handler->GetAccessingFabricIndex() == fabricIndex) |
| { |
| ChipLogProgress(InteractionModel, "Deleting expired ReadHandler for NodeId: " ChipLogFormatX64 ", FabricIndex: %u", |
| ChipLogValueX64(handler->GetInitiatorNodeId()), fabricIndex); |
| handler->Close(); |
| } |
| |
| return Loop::Continue; |
| }); |
| |
| for (auto * readClient = mpActiveReadClientList; readClient != nullptr; readClient = readClient->GetNextClient()) |
| { |
| if (readClient->GetFabricIndex() == fabricIndex) |
| { |
| ChipLogProgress(InteractionModel, "Fabric removed, deleting obsolete read client with FabricIndex: %u", fabricIndex); |
| readClient->Close(CHIP_ERROR_IM_FABRIC_DELETED, false); |
| } |
| } |
| |
| for (auto & handler : mWriteHandlers) |
| { |
| if (!(handler.IsFree()) && handler.GetAccessingFabricIndex() == fabricIndex) |
| { |
| ChipLogProgress(InteractionModel, "Fabric removed, deleting obsolete write handler with FabricIndex: %u", fabricIndex); |
| handler.Close(); |
| } |
| } |
| |
| // Applications may hold references to CommandHandler instances for async command processing. |
| // Therefore we can't forcible destroy CommandHandlers here. Their exchanges will get closed by |
| // the fabric removal, though, so they will fail when they try to actually send their command response |
| // and will close at that point. |
| } |
| } // namespace app |
| } // namespace chip |