// Copyright (c) 2012 The Chromium Authors. All rights reserved. // Use of this source code is governed by a BSD-style license that can be // found in the LICENSE file. #include "sql/connection.h" #include #include "base/files/file_path.h" #include "base/file_util.h" #include "base/logging.h" #include "base/metrics/histogram.h" #include "base/metrics/sparse_histogram.h" #include "base/strings/string_split.h" #include "base/strings/string_util.h" #include "base/strings/stringprintf.h" #include "base/strings/utf_string_conversions.h" #include "sql/statement.h" #include "third_party/sqlite/sqlite3.h" namespace { // Spin for up to a second waiting for the lock to clear when setting // up the database. // TODO(shess): Better story on this. http://crbug.com/56559 const int kBusyTimeoutSeconds = 1; class ScopedBusyTimeout { public: explicit ScopedBusyTimeout(sqlite3* db) : db_(db) { } ~ScopedBusyTimeout() { sqlite3_busy_timeout(db_, 0); } int SetTimeout(base::TimeDelta timeout) { DCHECK_LT(timeout.InMilliseconds(), INT_MAX); return sqlite3_busy_timeout(db_, static_cast(timeout.InMilliseconds())); } private: sqlite3* db_; }; // Helper to "safely" enable writable_schema. No error checking // because it is reasonable to just forge ahead in case of an error. // If turning it on fails, then most likely nothing will work, whereas // if turning it off fails, it only matters if some code attempts to // continue working with the database and tries to modify the // sqlite_master table (none of our code does this). class ScopedWritableSchema { public: explicit ScopedWritableSchema(sqlite3* db) : db_(db) { sqlite3_exec(db_, "PRAGMA writable_schema=1", NULL, NULL, NULL); } ~ScopedWritableSchema() { sqlite3_exec(db_, "PRAGMA writable_schema=0", NULL, NULL, NULL); } private: sqlite3* db_; }; } // namespace namespace sql { // static Connection::ErrorIgnorerCallback* Connection::current_ignorer_cb_ = NULL; // static bool Connection::ShouldIgnore(int error) { if (!current_ignorer_cb_) return false; return current_ignorer_cb_->Run(error); } // static void Connection::SetErrorIgnorer(Connection::ErrorIgnorerCallback* cb) { CHECK(current_ignorer_cb_ == NULL); current_ignorer_cb_ = cb; } // static void Connection::ResetErrorIgnorer() { CHECK(current_ignorer_cb_); current_ignorer_cb_ = NULL; } bool StatementID::operator<(const StatementID& other) const { if (number_ != other.number_) return number_ < other.number_; return strcmp(str_, other.str_) < 0; } Connection::StatementRef::StatementRef(Connection* connection, sqlite3_stmt* stmt, bool was_valid) : connection_(connection), stmt_(stmt), was_valid_(was_valid) { if (connection) connection_->StatementRefCreated(this); } Connection::StatementRef::~StatementRef() { if (connection_) connection_->StatementRefDeleted(this); Close(false); } void Connection::StatementRef::Close(bool forced) { if (stmt_) { // Call to AssertIOAllowed() cannot go at the beginning of the function // because Close() is called unconditionally from destructor to clean // connection_. And if this is inactive statement this won't cause any // disk access and destructor most probably will be called on thread // not allowing disk access. // TODO(paivanof@gmail.com): This should move to the beginning // of the function. http://crbug.com/136655. AssertIOAllowed(); sqlite3_finalize(stmt_); stmt_ = NULL; } connection_ = NULL; // The connection may be getting deleted. // Forced close is expected to happen from a statement error // handler. In that case maintain the sense of |was_valid_| which // previously held for this ref. was_valid_ = was_valid_ && forced; } Connection::Connection() : db_(NULL), page_size_(0), cache_size_(0), exclusive_locking_(false), transaction_nesting_(0), needs_rollback_(false), in_memory_(false), poisoned_(false) { } Connection::~Connection() { Close(); } bool Connection::Open(const base::FilePath& path) { if (!histogram_tag_.empty()) { int64 size_64 = 0; if (file_util::GetFileSize(path, &size_64)) { size_t sample = static_cast(size_64 / 1024); std::string full_histogram_name = "Sqlite.SizeKB." + histogram_tag_; base::HistogramBase* histogram = base::Histogram::FactoryGet( full_histogram_name, 1, 1000000, 50, base::HistogramBase::kUmaTargetedHistogramFlag); if (histogram) histogram->Add(sample); } } #if defined(OS_WIN) return OpenInternal(WideToUTF8(path.value())); #elif defined(OS_POSIX) return OpenInternal(path.value()); #endif } bool Connection::OpenInMemory() { in_memory_ = true; return OpenInternal(":memory:"); } void Connection::CloseInternal(bool forced) { // TODO(shess): Calling "PRAGMA journal_mode = DELETE" at this point // will delete the -journal file. For ChromiumOS or other more // embedded systems, this is probably not appropriate, whereas on // desktop it might make some sense. // sqlite3_close() needs all prepared statements to be finalized. // Release cached statements. statement_cache_.clear(); // With cached statements released, in-use statements will remain. // Closing the database while statements are in use is an API // violation, except for forced close (which happens from within a // statement's error handler). DCHECK(forced || open_statements_.empty()); // Deactivate any outstanding statements so sqlite3_close() works. for (StatementRefSet::iterator i = open_statements_.begin(); i != open_statements_.end(); ++i) (*i)->Close(forced); open_statements_.clear(); if (db_) { // Call to AssertIOAllowed() cannot go at the beginning of the function // because Close() must be called from destructor to clean // statement_cache_, it won't cause any disk access and it most probably // will happen on thread not allowing disk access. // TODO(paivanof@gmail.com): This should move to the beginning // of the function. http://crbug.com/136655. AssertIOAllowed(); // TODO(shess): Histogram for failure. sqlite3_close(db_); db_ = NULL; } } void Connection::Close() { // If the database was already closed by RazeAndClose(), then no // need to close again. Clear the |poisoned_| bit so that incorrect // API calls are caught. if (poisoned_) { poisoned_ = false; return; } CloseInternal(false); } void Connection::Preload() { AssertIOAllowed(); if (!db_) { DLOG_IF(FATAL, !poisoned_) << "Cannot preload null db"; return; } // A statement must be open for the preload command to work. If the meta // table doesn't exist, it probably means this is a new database and there // is nothing to preload (so it's OK we do nothing). if (!DoesTableExist("meta")) return; Statement dummy(GetUniqueStatement("SELECT * FROM meta")); if (!dummy.Step()) return; #if !defined(USE_SYSTEM_SQLITE) // This function is only defined in Chromium's version of sqlite. // Do not call it when using system sqlite. sqlite3_preload(db_); #endif } // Create an in-memory database with the existing database's page // size, then backup that database over the existing database. bool Connection::Raze() { AssertIOAllowed(); if (!db_) { DLOG_IF(FATAL, !poisoned_) << "Cannot raze null db"; return false; } if (transaction_nesting_ > 0) { DLOG(FATAL) << "Cannot raze within a transaction"; return false; } sql::Connection null_db; if (!null_db.OpenInMemory()) { DLOG(FATAL) << "Unable to open in-memory database."; return false; } if (page_size_) { // Enforce SQLite restrictions on |page_size_|. DCHECK(!(page_size_ & (page_size_ - 1))) << " page_size_ " << page_size_ << " is not a power of two."; const int kSqliteMaxPageSize = 32768; // from sqliteLimit.h DCHECK_LE(page_size_, kSqliteMaxPageSize); const std::string sql = base::StringPrintf("PRAGMA page_size=%d", page_size_); if (!null_db.Execute(sql.c_str())) return false; } #if defined(OS_ANDROID) // Android compiles with SQLITE_DEFAULT_AUTOVACUUM. Unfortunately, // in-memory databases do not respect this define. // TODO(shess): Figure out a way to set this without using platform // specific code. AFAICT from sqlite3.c, the only way to do it // would be to create an actual filesystem database, which is // unfortunate. if (!null_db.Execute("PRAGMA auto_vacuum = 1")) return false; #endif // The page size doesn't take effect until a database has pages, and // at this point the null database has none. Changing the schema // version will create the first page. This will not affect the // schema version in the resulting database, as SQLite's backup // implementation propagates the schema version from the original // connection to the new version of the database, incremented by one // so that other readers see the schema change and act accordingly. if (!null_db.Execute("PRAGMA schema_version = 1")) return false; // SQLite tracks the expected number of database pages in the first // page, and if it does not match the total retrieved from a // filesystem call, treats the database as corrupt. This situation // breaks almost all SQLite calls. "PRAGMA writable_schema" can be // used to hint to SQLite to soldier on in that case, specifically // for purposes of recovery. [See SQLITE_CORRUPT_BKPT case in // sqlite3.c lockBtree().] // TODO(shess): With this, "PRAGMA auto_vacuum" and "PRAGMA // page_size" can be used to query such a database. ScopedWritableSchema writable_schema(db_); sqlite3_backup* backup = sqlite3_backup_init(db_, "main", null_db.db_, "main"); if (!backup) { DLOG(FATAL) << "Unable to start sqlite3_backup()."; return false; } // -1 backs up the entire database. int rc = sqlite3_backup_step(backup, -1); int pages = sqlite3_backup_pagecount(backup); sqlite3_backup_finish(backup); // The destination database was locked. if (rc == SQLITE_BUSY) { return false; } // The entire database should have been backed up. if (rc != SQLITE_DONE) { DLOG(FATAL) << "Unable to copy entire null database."; return false; } // Exactly one page should have been backed up. If this breaks, // check this function to make sure assumptions aren't being broken. DCHECK_EQ(pages, 1); return true; } bool Connection::RazeWithTimout(base::TimeDelta timeout) { if (!db_) { DLOG_IF(FATAL, !poisoned_) << "Cannot raze null db"; return false; } ScopedBusyTimeout busy_timeout(db_); busy_timeout.SetTimeout(timeout); return Raze(); } bool Connection::RazeAndClose() { if (!db_) { DLOG_IF(FATAL, !poisoned_) << "Cannot raze null db"; return false; } // Raze() cannot run in a transaction. while (transaction_nesting_) { RollbackTransaction(); } bool result = Raze(); CloseInternal(true); // Mark the database so that future API calls fail appropriately, // but don't DCHECK (because after calling this function they are // expected to fail). poisoned_ = true; return result; } // TODO(shess): To the extent possible, figure out the optimal // ordering for these deletes which will prevent other connections // from seeing odd behavior. For instance, it may be necessary to // manually lock the main database file in a SQLite-compatible fashion // (to prevent other processes from opening it), then delete the // journal files, then delete the main database file. Another option // might be to lock the main database file and poison the header with // junk to prevent other processes from opening it successfully (like // Gears "SQLite poison 3" trick). // // static bool Connection::Delete(const base::FilePath& path) { base::ThreadRestrictions::AssertIOAllowed(); base::FilePath journal_path(path.value() + FILE_PATH_LITERAL("-journal")); base::FilePath wal_path(path.value() + FILE_PATH_LITERAL("-wal")); file_util::Delete(journal_path, false); file_util::Delete(wal_path, false); file_util::Delete(path, false); return !file_util::PathExists(journal_path) && !file_util::PathExists(wal_path) && !file_util::PathExists(path); } bool Connection::BeginTransaction() { if (needs_rollback_) { DCHECK_GT(transaction_nesting_, 0); // When we're going to rollback, fail on this begin and don't actually // mark us as entering the nested transaction. return false; } bool success = true; if (!transaction_nesting_) { needs_rollback_ = false; Statement begin(GetCachedStatement(SQL_FROM_HERE, "BEGIN TRANSACTION")); if (!begin.Run()) return false; } transaction_nesting_++; return success; } void Connection::RollbackTransaction() { if (!transaction_nesting_) { DLOG_IF(FATAL, !poisoned_) << "Rolling back a nonexistent transaction"; return; } transaction_nesting_--; if (transaction_nesting_ > 0) { // Mark the outermost transaction as needing rollback. needs_rollback_ = true; return; } DoRollback(); } bool Connection::CommitTransaction() { if (!transaction_nesting_) { DLOG_IF(FATAL, !poisoned_) << "Rolling back a nonexistent transaction"; return false; } transaction_nesting_--; if (transaction_nesting_ > 0) { // Mark any nested transactions as failing after we've already got one. return !needs_rollback_; } if (needs_rollback_) { DoRollback(); return false; } Statement commit(GetCachedStatement(SQL_FROM_HERE, "COMMIT")); return commit.Run(); } int Connection::ExecuteAndReturnErrorCode(const char* sql) { AssertIOAllowed(); if (!db_) { DLOG_IF(FATAL, !poisoned_) << "Illegal use of connection without a db"; return SQLITE_ERROR; } return sqlite3_exec(db_, sql, NULL, NULL, NULL); } bool Connection::Execute(const char* sql) { if (!db_) { DLOG_IF(FATAL, !poisoned_) << "Illegal use of connection without a db"; return false; } int error = ExecuteAndReturnErrorCode(sql); if (error != SQLITE_OK) error = OnSqliteError(error, NULL); // This needs to be a FATAL log because the error case of arriving here is // that there's a malformed SQL statement. This can arise in development if // a change alters the schema but not all queries adjust. This can happen // in production if the schema is corrupted. if (error == SQLITE_ERROR) DLOG(FATAL) << "SQL Error in " << sql << ", " << GetErrorMessage(); return error == SQLITE_OK; } bool Connection::ExecuteWithTimeout(const char* sql, base::TimeDelta timeout) { if (!db_) { DLOG_IF(FATAL, !poisoned_) << "Illegal use of connection without a db"; return false; } ScopedBusyTimeout busy_timeout(db_); busy_timeout.SetTimeout(timeout); return Execute(sql); } bool Connection::HasCachedStatement(const StatementID& id) const { return statement_cache_.find(id) != statement_cache_.end(); } scoped_refptr Connection::GetCachedStatement( const StatementID& id, const char* sql) { CachedStatementMap::iterator i = statement_cache_.find(id); if (i != statement_cache_.end()) { // Statement is in the cache. It should still be active (we're the only // one invalidating cached statements, and we'll remove it from the cache // if we do that. Make sure we reset it before giving out the cached one in // case it still has some stuff bound. DCHECK(i->second->is_valid()); sqlite3_reset(i->second->stmt()); return i->second; } scoped_refptr statement = GetUniqueStatement(sql); if (statement->is_valid()) statement_cache_[id] = statement; // Only cache valid statements. return statement; } scoped_refptr Connection::GetUniqueStatement( const char* sql) { AssertIOAllowed(); // Return inactive statement. if (!db_) return new StatementRef(NULL, NULL, poisoned_); sqlite3_stmt* stmt = NULL; int rc = sqlite3_prepare_v2(db_, sql, -1, &stmt, NULL); if (rc != SQLITE_OK) { // This is evidence of a syntax error in the incoming SQL. DLOG(FATAL) << "SQL compile error " << GetErrorMessage(); // It could also be database corruption. OnSqliteError(rc, NULL); return new StatementRef(NULL, NULL, false); } return new StatementRef(this, stmt, true); } scoped_refptr Connection::GetUntrackedStatement( const char* sql) const { // Return inactive statement. if (!db_) return new StatementRef(NULL, NULL, poisoned_); sqlite3_stmt* stmt = NULL; int rc = sqlite3_prepare_v2(db_, sql, -1, &stmt, NULL); if (rc != SQLITE_OK) { // This is evidence of a syntax error in the incoming SQL. DLOG(FATAL) << "SQL compile error " << GetErrorMessage(); return new StatementRef(NULL, NULL, false); } return new StatementRef(NULL, stmt, true); } bool Connection::IsSQLValid(const char* sql) { AssertIOAllowed(); if (!db_) { DLOG_IF(FATAL, !poisoned_) << "Illegal use of connection without a db"; return false; } sqlite3_stmt* stmt = NULL; if (sqlite3_prepare_v2(db_, sql, -1, &stmt, NULL) != SQLITE_OK) return false; sqlite3_finalize(stmt); return true; } bool Connection::DoesTableExist(const char* table_name) const { return DoesTableOrIndexExist(table_name, "table"); } bool Connection::DoesIndexExist(const char* index_name) const { return DoesTableOrIndexExist(index_name, "index"); } bool Connection::DoesTableOrIndexExist( const char* name, const char* type) const { const char* kSql = "SELECT name FROM sqlite_master WHERE type=? AND name=?"; Statement statement(GetUntrackedStatement(kSql)); statement.BindString(0, type); statement.BindString(1, name); return statement.Step(); // Table exists if any row was returned. } bool Connection::DoesColumnExist(const char* table_name, const char* column_name) const { std::string sql("PRAGMA TABLE_INFO("); sql.append(table_name); sql.append(")"); Statement statement(GetUntrackedStatement(sql.c_str())); while (statement.Step()) { if (!statement.ColumnString(1).compare(column_name)) return true; } return false; } int64 Connection::GetLastInsertRowId() const { if (!db_) { DLOG_IF(FATAL, !poisoned_) << "Illegal use of connection without a db"; return 0; } return sqlite3_last_insert_rowid(db_); } int Connection::GetLastChangeCount() const { if (!db_) { DLOG_IF(FATAL, !poisoned_) << "Illegal use of connection without a db"; return 0; } return sqlite3_changes(db_); } int Connection::GetErrorCode() const { if (!db_) return SQLITE_ERROR; return sqlite3_errcode(db_); } int Connection::GetLastErrno() const { if (!db_) return -1; int err = 0; if (SQLITE_OK != sqlite3_file_control(db_, NULL, SQLITE_LAST_ERRNO, &err)) return -2; return err; } const char* Connection::GetErrorMessage() const { if (!db_) return "sql::Connection has no connection."; return sqlite3_errmsg(db_); } bool Connection::OpenInternal(const std::string& file_name) { AssertIOAllowed(); if (db_) { DLOG(FATAL) << "sql::Connection is already open."; return false; } // If |poisoned_| is set, it means an error handler called // RazeAndClose(). Until regular Close() is called, the caller // should be treating the database as open, but is_open() currently // only considers the sqlite3 handle's state. // TODO(shess): Revise is_open() to consider poisoned_, and review // to see if any non-testing code even depends on it. DLOG_IF(FATAL, poisoned_) << "sql::Connection is already open."; int err = sqlite3_open(file_name.c_str(), &db_); if (err != SQLITE_OK) { // Histogram failures specific to initial open for debugging // purposes. UMA_HISTOGRAM_ENUMERATION("Sqlite.OpenFailure", err & 0xff, 50); OnSqliteError(err, NULL); Close(); db_ = NULL; return false; } // SQLite uses a lookaside buffer to improve performance of small mallocs. // Chromium already depends on small mallocs being efficient, so we disable // this to avoid the extra memory overhead. // This must be called immediatly after opening the database before any SQL // statements are run. sqlite3_db_config(db_, SQLITE_DBCONFIG_LOOKASIDE, NULL, 0, 0); // sqlite3_open() does not actually read the database file (unless a // hot journal is found). Successfully executing this pragma on an // existing database requires a valid header on page 1. // TODO(shess): For now, just probing to see what the lay of the // land is. If it's mostly SQLITE_NOTADB, then the database should // be razed. err = ExecuteAndReturnErrorCode("PRAGMA auto_vacuum"); if (err != SQLITE_OK) UMA_HISTOGRAM_ENUMERATION("Sqlite.OpenProbeFailure", err & 0xff, 50); // Enable extended result codes to provide more color on I/O errors. // Not having extended result codes is not a fatal problem, as // Chromium code does not attempt to handle I/O errors anyhow. The // current implementation always returns SQLITE_OK, the DCHECK is to // quickly notify someone if SQLite changes. err = sqlite3_extended_result_codes(db_, 1); DCHECK_EQ(err, SQLITE_OK) << "Could not enable extended result codes"; // If indicated, lock up the database before doing anything else, so // that the following code doesn't have to deal with locking. // TODO(shess): This code is brittle. Find the cases where code // doesn't request |exclusive_locking_| and audit that it does the // right thing with SQLITE_BUSY, and that it doesn't make // assumptions about who might change things in the database. // http://crbug.com/56559 if (exclusive_locking_) { // TODO(shess): This should probably be a failure. Code which // requests exclusive locking but doesn't get it is almost certain // to be ill-tested. ignore_result(Execute("PRAGMA locking_mode=EXCLUSIVE")); } // http://www.sqlite.org/pragma.html#pragma_journal_mode // DELETE (default) - delete -journal file to commit. // TRUNCATE - truncate -journal file to commit. // PERSIST - zero out header of -journal file to commit. // journal_size_limit provides size to trim to in PERSIST. // TODO(shess): Figure out if PERSIST and journal_size_limit really // matter. In theory, it keeps pages pre-allocated, so if // transactions usually fit, it should be faster. ignore_result(Execute("PRAGMA journal_mode = PERSIST")); ignore_result(Execute("PRAGMA journal_size_limit = 16384")); const base::TimeDelta kBusyTimeout = base::TimeDelta::FromSeconds(kBusyTimeoutSeconds); if (page_size_ != 0) { // Enforce SQLite restrictions on |page_size_|. DCHECK(!(page_size_ & (page_size_ - 1))) << " page_size_ " << page_size_ << " is not a power of two."; const int kSqliteMaxPageSize = 32768; // from sqliteLimit.h DCHECK_LE(page_size_, kSqliteMaxPageSize); const std::string sql = base::StringPrintf("PRAGMA page_size=%d", page_size_); ignore_result(ExecuteWithTimeout(sql.c_str(), kBusyTimeout)); } if (cache_size_ != 0) { const std::string sql = base::StringPrintf("PRAGMA cache_size=%d", cache_size_); ignore_result(ExecuteWithTimeout(sql.c_str(), kBusyTimeout)); } if (!ExecuteWithTimeout("PRAGMA secure_delete=ON", kBusyTimeout)) { Close(); return false; } return true; } void Connection::DoRollback() { Statement rollback(GetCachedStatement(SQL_FROM_HERE, "ROLLBACK")); rollback.Run(); needs_rollback_ = false; } void Connection::StatementRefCreated(StatementRef* ref) { DCHECK(open_statements_.find(ref) == open_statements_.end()); open_statements_.insert(ref); } void Connection::StatementRefDeleted(StatementRef* ref) { StatementRefSet::iterator i = open_statements_.find(ref); if (i == open_statements_.end()) DLOG(FATAL) << "Could not find statement"; else open_statements_.erase(i); } void Connection::AddTaggedHistogram(const std::string& name, size_t sample) const { if (histogram_tag_.empty()) return; // TODO(shess): The histogram macros create a bit of static storage // for caching the histogram object. This code shouldn't execute // often enough for such caching to be crucial. If it becomes an // issue, the object could be cached alongside histogram_prefix_. std::string full_histogram_name = name + "." + histogram_tag_; base::HistogramBase* histogram = base::SparseHistogram::FactoryGet( full_histogram_name, base::HistogramBase::kUmaTargetedHistogramFlag); if (histogram) histogram->Add(sample); } int Connection::OnSqliteError(int err, sql::Statement *stmt) { UMA_HISTOGRAM_SPARSE_SLOWLY("Sqlite.Error", err); AddTaggedHistogram("Sqlite.Error", err); // Always log the error. LOG(ERROR) << "sqlite error " << err << ", errno " << GetLastErrno() << ": " << GetErrorMessage(); if (!error_callback_.is_null()) { error_callback_.Run(err, stmt); return err; } // The default handling is to assert on debug and to ignore on release. if (!ShouldIgnore(err)) DLOG(FATAL) << GetErrorMessage(); return err; } // TODO(shess): Allow specifying integrity_check versus quick_check. // TODO(shess): Allow specifying maximum results (default 100 lines). bool Connection::IntegrityCheck(std::vector* messages) { messages->clear(); // This has the side effect of setting SQLITE_RecoveryMode, which // allows SQLite to process through certain cases of corruption. // Failing to set this pragma probably means that the database is // beyond recovery. const char kWritableSchema[] = "PRAGMA writable_schema = ON"; if (!Execute(kWritableSchema)) return false; bool ret = false; { const char kSql[] = "PRAGMA integrity_check"; sql::Statement stmt(GetUniqueStatement(kSql)); // The pragma appears to return all results (up to 100 by default) // as a single string. This doesn't appear to be an API contract, // it could return separate lines, so loop _and_ split. while (stmt.Step()) { std::string result(stmt.ColumnString(0)); base::SplitString(result, '\n', messages); } ret = stmt.Succeeded(); } // Best effort to put things back as they were before. const char kNoWritableSchema[] = "PRAGMA writable_schema = OFF"; ignore_result(Execute(kNoWritableSchema)); return ret; } } // namespace sql