/* * Copyright (C) 2007, 2008, 2013 Apple Inc. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of Apple Inc. ("Apple") nor the names of * its contributors may be used to endorse or promote products derived * from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY APPLE AND ITS CONTRIBUTORS "AS IS" AND ANY * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE * DISCLAIMED. IN NO EVENT SHALL APPLE OR ITS CONTRIBUTORS BE LIABLE FOR ANY * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "config.h" #include "SQLTransactionBackend.h" #include "Database.h" #include "DatabaseAuthorizer.h" #include "DatabaseContext.h" #include "DatabaseThread.h" #include "DatabaseTracker.h" #include "Logging.h" #include "OriginLock.h" #include "SQLError.h" #include "SQLStatement.h" #include "SQLStatementCallback.h" #include "SQLStatementErrorCallback.h" #include "SQLTransaction.h" #include "SQLTransactionCoordinator.h" #include "SQLValue.h" #include "SQLiteTransaction.h" #include <wtf/StdLibExtras.h> #include <wtf/text/WTFString.h> // How does a SQLTransaction work? // ============================== // The SQLTransaction is a state machine that executes a series of states / steps. // // The work of the transaction states are defined in section of 4.3.2 of the // webdatabase spec: http://dev.w3.org/html5/webdatabase/#processing-model // // the State Transition Graph at a glance: // ====================================== // // Backend . Frontend // (works with SQLiteDatabase) . (works with Script) // =========================== . =================== // . // 1. Idle . // v . // 2. AcquireLock . // v . // 3. OpenTransactionAndPreflight ------------------------------------------. // | . | // `-------------------------------> 8. DeliverTransactionCallback --. | // . | v v // ,-------------------------------------' 9. DeliverTransactionErrorCallback + // | . ^ ^ ^ | // v . | | | | // 4. RunStatements -----------------------------------------------------' | | | // | ^ ^ | ^ | . | | | // |--------' | | | `------------> 10. DeliverStatementCallback +-----' | | // | | | `---------------------------------------------' | | // | | `-----------------> 11. DeliverQuotaIncreaseCallback + | | // | `-----------------------------------------------------' | | // v . | | // 5. PostflightAndCommit --+--------------------------------------------------' | // |----------> 12. DeliverSuccessCallback + | // ,--------------------' . | | // v . | | // 6. CleanupAndTerminate <-----------------------------------------' | // v ^ . | // 0. End | . | // | . | // 7: CleanupAfterTransactionErrorCallback <----------------------------' // . // // the States and State Transitions: // ================================ // 0. SQLTransactionState::End // - the end state. // // 1. SQLTransactionState::Idle // - placeholder state while waiting on frontend/backend, etc. See comment on // "State transitions between SQLTransaction and SQLTransactionBackend" // below. // // 2. SQLTransactionState::AcquireLock (runs in backend) // - this is the start state. // - acquire the "lock". // - on "lock" acquisition, goto SQLTransactionState::OpenTransactionAndPreflight. // // 3. SQLTransactionState::openTransactionAndPreflight (runs in backend) // - Sets up an SQLiteTransaction. // - begin the SQLiteTransaction. // - call the SQLTransactionWrapper preflight if available. // - schedule script callback. // - on error, goto SQLTransactionState::DeliverTransactionErrorCallback. // - goto SQLTransactionState::DeliverTransactionCallback. // // 4. SQLTransactionState::DeliverTransactionCallback (runs in frontend) // - invoke the script function callback() if available. // - on error, goto SQLTransactionState::DeliverTransactionErrorCallback. // - goto SQLTransactionState::RunStatements. // // 5. SQLTransactionState::DeliverTransactionErrorCallback (runs in frontend) // - invoke the script function errorCallback if available. // - goto SQLTransactionState::CleanupAfterTransactionErrorCallback. // // 6. SQLTransactionState::RunStatements (runs in backend) // - while there are statements { // - run a statement. // - if statementCallback is available, goto SQLTransactionState::DeliverStatementCallback. // - on error, // goto SQLTransactionState::DeliverQuotaIncreaseCallback, or // goto SQLTransactionState::DeliverStatementCallback, or // goto SQLTransactionState::deliverTransactionErrorCallback. // } // - goto SQLTransactionState::PostflightAndCommit. // // 7. SQLTransactionState::DeliverStatementCallback (runs in frontend) // - invoke script statement callback (assume available). // - on error, goto SQLTransactionState::DeliverTransactionErrorCallback. // - goto SQLTransactionState::RunStatements. // // 8. SQLTransactionState::DeliverQuotaIncreaseCallback (runs in frontend) // - give client a chance to increase the quota. // - goto SQLTransactionState::RunStatements. // // 9. SQLTransactionState::PostflightAndCommit (runs in backend) // - call the SQLTransactionWrapper postflight if available. // - commit the SQLiteTansaction. // - on error, goto SQLTransactionState::DeliverTransactionErrorCallback. // - if successCallback is available, goto SQLTransactionState::DeliverSuccessCallback. // else goto SQLTransactionState::CleanupAndTerminate. // // 10. SQLTransactionState::DeliverSuccessCallback (runs in frontend) // - invoke the script function successCallback() if available. // - goto SQLTransactionState::CleanupAndTerminate. // // 11. SQLTransactionState::CleanupAndTerminate (runs in backend) // - stop and clear the SQLiteTransaction. // - release the "lock". // - goto SQLTransactionState::End. // // 12. SQLTransactionState::CleanupAfterTransactionErrorCallback (runs in backend) // - rollback the SQLiteTransaction. // - goto SQLTransactionState::CleanupAndTerminate. // // State transitions between SQLTransaction and SQLTransactionBackend // ================================================================== // As shown above, there are state transitions that crosses the boundary between // the frontend and backend. For example, // // OpenTransactionAndPreflight (state 3 in the backend) // transitions to DeliverTransactionCallback (state 8 in the frontend), // which in turn transitions to RunStatements (state 4 in the backend). // // This cross boundary transition is done by posting transition requests to the // other side and letting the other side's state machine execute the state // transition in the appropriate thread (i.e. the script thread for the frontend, // and the database thread for the backend). // // Logically, the state transitions work as shown in the graph above. But // physically, the transition mechanism uses the Idle state (both in the frontend // and backend) as a waiting state for further activity. For example, taking a // closer look at the 3 state transition example above, what actually happens // is as follows: // // Step 1: // ====== // In the frontend thread: // - waiting quietly is Idle. Not doing any work. // // In the backend: // - is in OpenTransactionAndPreflight, and doing its work. // - when done, it transits to the backend DeliverTransactionCallback. // - the backend DeliverTransactionCallback sends a request to the frontend // to transit to DeliverTransactionCallback, and then itself transits to // Idle. // // Step 2: // ====== // In the frontend thread: // - transits to DeliverTransactionCallback and does its work. // - when done, it transits to the frontend RunStatements. // - the frontend RunStatements sends a request to the backend to transit // to RunStatements, and then itself transits to Idle. // // In the backend: // - waiting quietly in Idle. // // Step 3: // ====== // In the frontend thread: // - waiting quietly is Idle. Not doing any work. // // In the backend: // - transits to RunStatements, and does its work. // ... // // So, when the frontend or backend are not active, they will park themselves in // their Idle states. This means their m_nextState is set to Idle, but they never // actually run the corresponding state function. Note: for both the frontend and // backend, the state function for Idle is unreachableState(). // // The states that send a request to their peer across the front/back boundary // are implemented with just 2 functions: SQLTransaction::sendToBackendState() // and SQLTransactionBackend::sendToFrontendState(). These state functions do // nothing but sends a request to the other side to transit to the current // state (indicated by m_nextState), and then transits itself to the Idle state // to wait for further action. // The Life-Cycle of a SQLTransaction i.e. Who's keeping the SQLTransaction alive? // ============================================================================== // The RefPtr chain goes something like this: // // At birth (in DatabaseBackend::runTransaction()): // ==================================================== // DatabaseBackend // Deque<RefPtr<SQLTransactionBackend>> m_transactionQueue points to ... // --> SQLTransactionBackend // RefPtr<SQLTransaction> m_frontend points to ... // --> SQLTransaction // RefPtr<SQLTransactionBackend> m_backend points to ... // --> SQLTransactionBackend // which is a circular reference. // // Note: there's a circular reference between the SQLTransaction front-end and // back-end. This circular reference is established in the constructor of the // SQLTransactionBackend. The circular reference will be broken by calling // doCleanup() to nullify m_frontend. This is done at the end of the transaction's // clean up state (i.e. when the transaction should no longer be in use thereafter), // or if the database was interrupted. See comments on "What happens if a transaction // is interrupted?" below for details. // // After scheduling the transaction with the DatabaseThread (DatabaseBackend::scheduleTransaction()): // ====================================================================================================== // DatabaseThread // MessageQueue<DatabaseTask> m_queue points to ... // --> DatabaseTransactionTask // RefPtr<SQLTransactionBackend> m_transaction points to ... // --> SQLTransactionBackend // RefPtr<SQLTransaction> m_frontend points to ... // --> SQLTransaction // RefPtr<SQLTransactionBackend> m_backend points to ... // --> SQLTransactionBackend // which is a circular reference. // // When executing the transaction (in DatabaseThread::databaseThread()): // ==================================================================== // std::unique_ptr<DatabaseTask> task; // points to ... // --> DatabaseTransactionTask // RefPtr<SQLTransactionBackend> m_transaction points to ... // --> SQLTransactionBackend // RefPtr<SQLTransaction> m_frontend; // --> SQLTransaction // RefPtr<SQLTransactionBackend> m_backend points to ... // --> SQLTransactionBackend // which is a circular reference. // // At the end of cleanupAndTerminate(): // =================================== // At the end of the cleanup state, the SQLTransactionBackend::m_frontend is nullified. // If by then, a JSObject wrapper is referring to the SQLTransaction, then the reference // chain looks like this: // // JSObjectWrapper // --> SQLTransaction // in RefPtr<SQLTransactionBackend> m_backend points to ... // --> SQLTransactionBackend // which no longer points back to its SQLTransaction. // // When the GC collects the corresponding JSObject, the above chain will be cleaned up // and deleted. // // If there is no JSObject wrapper referring to the SQLTransaction when the cleanup // states nullify SQLTransactionBackend::m_frontend, the SQLTransaction will deleted then. // However, there will still be a DatabaseTask pointing to the SQLTransactionBackend (see // the "When executing the transaction" chain above). This will keep the // SQLTransactionBackend alive until DatabaseThread::databaseThread() releases its // task std::unique_ptr. // // What happens if a transaction is interrupted? // ============================================ // If the transaction is interrupted half way, it won't get to run to state // CleanupAndTerminate, and hence, would not have called SQLTransactionBackend's // doCleanup(). doCleanup() is where we nullify SQLTransactionBackend::m_frontend // to break the reference cycle between the frontend and backend. Hence, we need // to cleanup the transaction by other means. // // Note: calling SQLTransactionBackend::notifyDatabaseThreadIsShuttingDown() // is effectively the same as calling SQLTransactionBackend::doClean(). // // In terms of who needs to call doCleanup(), there are 5 phases in the // SQLTransactionBackend life-cycle. These are the phases and how the clean // up is done: // // Phase 1. After Birth, before scheduling // // - To clean up, DatabaseThread::databaseThread() will call // DatabaseBackend::close() during its shutdown. // - DatabaseBackend::close() will iterate // DatabaseBackend::m_transactionQueue and call // notifyDatabaseThreadIsShuttingDown() on each transaction there. // // Phase 2. After scheduling, before state AcquireLock // // - If the interruption occures before the DatabaseTransactionTask is // scheduled in DatabaseThread::m_queue but hasn't gotten to execute // (i.e. DatabaseTransactionTask::performTask() has not been called), // then the DatabaseTransactionTask may get destructed before it ever // gets to execute. // - To clean up, the destructor will check if the task's m_wasExecuted is // set. If not, it will call notifyDatabaseThreadIsShuttingDown() on // the task's transaction. // // Phase 3. After state AcquireLock, before "lockAcquired" // // - In this phase, the transaction would have been added to the // SQLTransactionCoordinator's CoordinationInfo's pendingTransactions. // - To clean up, during shutdown, DatabaseThread::databaseThread() calls // SQLTransactionCoordinator::shutdown(), which calls // notifyDatabaseThreadIsShuttingDown(). // // Phase 4: After "lockAcquired", before state CleanupAndTerminate // // - In this phase, the transaction would have been added either to the // SQLTransactionCoordinator's CoordinationInfo's activeWriteTransaction // or activeReadTransactions. // - To clean up, during shutdown, DatabaseThread::databaseThread() calls // SQLTransactionCoordinator::shutdown(), which calls // notifyDatabaseThreadIsShuttingDown(). // // Phase 5: After state CleanupAndTerminate // // - This is how a transaction ends normally. // - state CleanupAndTerminate calls doCleanup(). namespace WebCore { SQLTransactionBackend::SQLTransactionBackend(SQLTransaction& frontend) : m_frontend(frontend) { m_requestedState = SQLTransactionState::AcquireLock; } SQLTransactionBackend::~SQLTransactionBackend() { ASSERT(!m_frontend.m_sqliteTransaction); } void SQLTransactionBackend::doCleanup() { ASSERT(currentThread() == m_frontend.database().databaseThread().getThreadID()); m_frontend.releaseOriginLockIfNeeded(); LockHolder locker(m_frontend.m_statementMutex); m_frontend.m_statementQueue.clear(); if (m_frontend.m_sqliteTransaction) { // In the event we got here because of an interruption or error (i.e. if // the transaction is in progress), we should roll it back here. Clearing // m_sqliteTransaction invokes SQLiteTransaction's destructor which does // just that. We might as well do this unconditionally and free up its // resources because we're already terminating. m_frontend.m_sqliteTransaction = nullptr; } // Release the lock on this database if (m_frontend.m_lockAcquired) m_frontend.m_database->transactionCoordinator()->releaseLock(m_frontend); // Do some aggresive clean up here except for m_database. // // We can't clear m_database here because the frontend may asynchronously // invoke SQLTransactionBackend::requestTransitToState(), and that function // uses m_database to schedule a state transition. This may occur because // the frontend (being in another thread) may already be on the way to // requesting our next state before it detects an interruption. // // There is no harm in letting it finish making the request. It'll set // m_requestedState, but we won't execute a transition to that state because // we've already shut down the transaction. // // We also can't clear m_currentStatementBackend and m_transactionError. // m_currentStatementBackend may be accessed asynchronously by the // frontend's deliverStatementCallback() state. Similarly, // m_transactionError may be accessed by deliverTransactionErrorCallback(). // This occurs if requests for transition to those states have already been // registered with the frontend just prior to a clean up request arriving. // // So instead, let our destructor handle their clean up since this // SQLTransactionBackend is guaranteed to not destruct until the frontend // is also destructing. m_frontend.m_wrapper = nullptr; } SQLTransactionBackend::StateFunction SQLTransactionBackend::stateFunctionFor(SQLTransactionState state) { static const StateFunction stateFunctions[] = { &SQLTransactionBackend::unreachableState, // 0. end &SQLTransactionBackend::unreachableState, // 1. idle &SQLTransactionBackend::acquireLock, // 2. &SQLTransactionBackend::openTransactionAndPreflight, // 3. &SQLTransactionBackend::runStatements, // 4. &SQLTransactionBackend::unreachableState, // 5. postflightAndCommit &SQLTransactionBackend::cleanupAndTerminate, // 6. &SQLTransactionBackend::cleanupAfterTransactionErrorCallback, // 7. &SQLTransactionBackend::unreachableState, // 8. deliverTransactionCallback &SQLTransactionBackend::unreachableState, // 9. deliverTransactionErrorCallback &SQLTransactionBackend::unreachableState, // 10. deliverStatementCallback &SQLTransactionBackend::unreachableState, // 11. deliverQuotaIncreaseCallback &SQLTransactionBackend::unreachableState // 12. deliverSuccessCallback }; ASSERT(WTF_ARRAY_LENGTH(stateFunctions) == static_cast<int>(SQLTransactionState::NumberOfStates)); ASSERT(state < SQLTransactionState::NumberOfStates); return stateFunctions[static_cast<int>(state)]; } void SQLTransactionBackend::computeNextStateAndCleanupIfNeeded() { // Only honor the requested state transition if we're not supposed to be // cleaning up and shutting down: if (m_frontend.m_database->opened()) { setStateToRequestedState(); ASSERT(m_nextState == SQLTransactionState::AcquireLock || m_nextState == SQLTransactionState::OpenTransactionAndPreflight || m_nextState == SQLTransactionState::RunStatements || m_nextState == SQLTransactionState::PostflightAndCommit || m_nextState == SQLTransactionState::CleanupAndTerminate || m_nextState == SQLTransactionState::CleanupAfterTransactionErrorCallback); LOG(StorageAPI, "State %s\n", nameForSQLTransactionState(m_nextState)); return; } // If we get here, then we should be shutting down. Do clean up if needed: if (m_nextState == SQLTransactionState::End) return; m_nextState = SQLTransactionState::End; // If the database was stopped, don't do anything and cancel queued work LOG(StorageAPI, "Database was stopped or interrupted - cancelling work for this transaction"); // The current SQLite transaction should be stopped, as well if (m_frontend.m_sqliteTransaction) { m_frontend.m_sqliteTransaction->stop(); m_frontend.m_sqliteTransaction = nullptr; } // Terminate the frontend state machine. This also gets the frontend to // call computeNextStateAndCleanupIfNeeded() and clear its wrappers // if needed. m_frontend.requestTransitToState(SQLTransactionState::End); // Redirect to the end state to abort, clean up, and end the transaction. doCleanup(); } void SQLTransactionBackend::notifyDatabaseThreadIsShuttingDown() { ASSERT(currentThread() == m_frontend.database().databaseThread().getThreadID()); // If the transaction is in progress, we should roll it back here, since this // is our last opportunity to do something related to this transaction on the // DB thread. Amongst other work, doCleanup() will clear m_sqliteTransaction // which invokes SQLiteTransaction's destructor, which will do the roll back // if necessary. doCleanup(); } void SQLTransactionBackend::acquireLock() { m_frontend.acquireLock(); } void SQLTransactionBackend::openTransactionAndPreflight() { m_frontend.openTransactionAndPreflight(); } void SQLTransactionBackend::runStatements() { m_frontend.runStatements(); } void SQLTransactionBackend::cleanupAndTerminate() { m_frontend.cleanupAndTerminate(); } void SQLTransactionBackend::cleanupAfterTransactionErrorCallback() { m_frontend.cleanupAfterTransactionErrorCallback(); } // requestTransitToState() can be called from the frontend. Hence, it should // NOT be modifying SQLTransactionBackend in general. The only safe field to // modify is m_requestedState which is meant for this purpose. void SQLTransactionBackend::requestTransitToState(SQLTransactionState nextState) { LOG(StorageAPI, "Scheduling %s for transaction %p\n", nameForSQLTransactionState(nextState), this); m_requestedState = nextState; ASSERT(m_requestedState != SQLTransactionState::End); m_frontend.m_database->scheduleTransactionStep(m_frontend); } // This state function is used as a stub function to plug unimplemented states // in the state dispatch table. They are unimplemented because they should // never be reached in the course of correct execution. void SQLTransactionBackend::unreachableState() { ASSERT_NOT_REACHED(); } } // namespace WebCore