/** * Copyright (C) 2018-present MongoDB, Inc. * * This program is free software: you can redistribute it and/or modify * it under the terms of the Server Side Public License, version 1, * as published by MongoDB, Inc. * * This program is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * Server Side Public License for more details. * * You should have received a copy of the Server Side Public License * along with this program. If not, see * . * * As a special exception, the copyright holders give permission to link the * code of portions of this program with the OpenSSL library under certain * conditions as described in each individual source file and distribute * linked combinations including the program with the OpenSSL library. You * must comply with the Server Side Public License in all respects for * all of the code used other than as permitted herein. If you modify file(s) * with this exception, you may extend this exception to your version of the * file(s), but you are not obligated to do so. If you do not wish to do so, * delete this exception statement from your version. If you delete this * exception statement from all source files in the program, then also delete * it in the license file. */ #include "mongo/client/dbclient_cursor.h" #include #include "mongo/client/connpool.h" #include "mongo/db/client.h" #include "mongo/db/dbmessage.h" #include "mongo/db/namespace_string.h" #include "mongo/db/pipeline/aggregate_command_gen.h" #include "mongo/db/pipeline/aggregation_request_helper.h" #include "mongo/db/query/cursor_response.h" #include "mongo/db/query/getmore_command_gen.h" #include "mongo/db/query/query_request_helper.h" #include "mongo/logv2/log.h" #include "mongo/rpc/factory.h" #include "mongo/rpc/get_status_from_command_result.h" #include "mongo/rpc/metadata.h" #include "mongo/s/stale_exception.h" #include "mongo/util/debug_util.h" #include "mongo/util/destructor_guard.h" #include "mongo/util/exit.h" #include "mongo/util/scopeguard.h" #define MONGO_LOGV2_DEFAULT_COMPONENT ::mongo::logv2::LogComponent::kNetwork namespace mongo { using std::endl; using std::string; using std::unique_ptr; using std::vector; namespace { BSONObj addMetadata(DBClientBase* client, BSONObj command) { if (client->getRequestMetadataWriter()) { BSONObjBuilder builder(command); auto opCtx = (haveClient() ? cc().getOperationContext() : nullptr); uassertStatusOK(client->getRequestMetadataWriter()(opCtx, &builder)); return builder.obj(); } else { return command; } } Message assembleCommandRequest(DBClientBase* client, const DatabaseName& dbName, BSONObj commandObj, const ReadPreferenceSetting& readPref) { // Add the $readPreference field to the request. { BSONObjBuilder builder{commandObj}; readPref.toContainingBSON(&builder); commandObj = builder.obj(); } commandObj = addMetadata(client, std::move(commandObj)); auto opMsgRequest = OpMsgRequestBuilder::create(dbName, commandObj); return opMsgRequest.serialize(); } } // namespace Message DBClientCursor::assembleInit() { if (_cursorId) { return assembleGetMore(); } // We haven't gotten a cursorId yet so we need to issue the initial find command. invariant(_findRequest); BSONObj findCmd = _findRequest->toBSON(BSONObj()); return assembleCommandRequest(_client, _ns.dbName(), std::move(findCmd), _readPref); } Message DBClientCursor::assembleGetMore() { invariant(_cursorId); auto getMoreRequest = GetMoreCommandRequest(_cursorId, _ns.coll().toString()); getMoreRequest.setBatchSize( boost::make_optional(_batchSize != 0, static_cast(_batchSize))); getMoreRequest.setMaxTimeMS(boost::make_optional( tailableAwaitData(), static_cast(durationCount(_awaitDataTimeout)))); if (_term) { getMoreRequest.setTerm(static_cast(*_term)); } getMoreRequest.setLastKnownCommittedOpTime(_lastKnownCommittedOpTime); auto msg = assembleCommandRequest(_client, _ns.dbName(), getMoreRequest.toBSON({}), _readPref); // Set the exhaust flag if needed. if (_isExhaust) { OpMsg::setFlag(&msg, OpMsg::kExhaustSupported); } return msg; } bool DBClientCursor::init() { invariant(!_connectionHasPendingReplies); Message toSend = assembleInit(); verify(_client); Message reply; try { _client->call(toSend, reply, &_originalHost); } catch (const DBException&) { // log msg temp? LOGV2(20127, "DBClientCursor::init call() failed"); // We always want to throw on network exceptions. throw; } if (reply.empty()) { // log msg temp? LOGV2(20128, "DBClientCursor::init message from call() was empty"); return false; } dataReceived(reply); _isInitialized = true; return true; } void DBClientCursor::requestMore() { // For exhaust queries, once the stream has been initiated we get data blasted to us // from the remote server, without a need to send any more 'getMore' requests. if (_isExhaust && _connectionHasPendingReplies) { return exhaustReceiveMore(); } invariant(!_connectionHasPendingReplies); verify(_cursorId && _batch.pos == _batch.objs.size()); auto doRequestMore = [&] { Message toSend = assembleGetMore(); Message response; _client->call(toSend, response); dataReceived(response); }; if (_client) return doRequestMore(); invariant(_scopedHost.size()); DBClientBase::withConnection_do_not_use(_scopedHost, [&](DBClientBase* conn) { ON_BLOCK_EXIT([&, origClient = _client] { _client = origClient; }); _client = conn; doRequestMore(); }); } /** * For exhaust cursors, the server just blasts data at us. The end of a stream is marked with a * cursor id of 0. */ void DBClientCursor::exhaustReceiveMore() { verify(_cursorId); verify(_batch.pos == _batch.objs.size()); Message response; verify(_client); uassertStatusOK( _client->recv(response, _lastRequestId).withContext("recv failed while exhausting cursor")); dataReceived(response); } BSONObj DBClientCursor::commandDataReceived(const Message& reply) { int op = reply.operation(); invariant(op == opReply || op == dbMsg); // Check if the reply indicates that it is part of an exhaust stream. const auto isExhaustReply = OpMsg::isFlagSet(reply, OpMsg::kMoreToCome); _connectionHasPendingReplies = isExhaustReply; if (isExhaustReply) { _lastRequestId = reply.header().getId(); } auto commandReply = _client->parseCommandReplyMessage(_client->getServerAddress(), reply); auto commandStatus = getStatusFromCommandResult(commandReply->getCommandReply()); if (commandStatus == ErrorCodes::StaleConfig) { uassertStatusOK( commandStatus.withContext("stale config in DBClientCursor::dataReceived()")); } else if (!commandStatus.isOK()) { _wasError = true; } return commandReply->getCommandReply().getOwned(); } void DBClientCursor::dataReceived(const Message& reply, bool& retry, string& host) { _batch.objs.clear(); _batch.pos = 0; const auto replyObj = commandDataReceived(reply); _cursorId = 0; // Don't try to kill cursor if we get back an error. auto cr = uassertStatusOK(CursorResponse::parseFromBSON(replyObj, nullptr, _ns.tenantId())); _cursorId = cr.getCursorId(); uassert(50935, "Received a getMore response with a cursor id of 0 and the moreToCome flag set.", !(_connectionHasPendingReplies && _cursorId == 0)); _ns = cr.getNSS(); // find command can change the ns to use for getMores. // Store the resume token, if we got one. _postBatchResumeToken = cr.getPostBatchResumeToken(); _batch.objs = cr.releaseBatch(); if (replyObj.hasField(LogicalTime::kOperationTimeFieldName)) { _operationTime = LogicalTime::fromOperationTime(replyObj).asTimestamp(); } } /** If true, safe to call next(). Requests more from server if necessary. */ bool DBClientCursor::more() { invariant(_isInitialized); if (!_putBack.empty()) return true; if (_batch.pos < _batch.objs.size()) return true; if (_cursorId == 0) return false; requestMore(); return _batch.pos < _batch.objs.size(); } BSONObj DBClientCursor::next() { invariant(_isInitialized); if (!_putBack.empty()) { BSONObj ret = _putBack.top(); _putBack.pop(); return ret; } uassert( 13422, "DBClientCursor next() called but more() is false", _batch.pos < _batch.objs.size()); return std::move(_batch.objs[_batch.pos++]); } BSONObj DBClientCursor::nextSafe() { BSONObj o = next(); // Only convert legacy errors ($err) to exceptions. Otherwise, just return the response and the // caller will interpret it as a command error. if (_wasError && strcmp(o.firstElementFieldName(), "$err") == 0) { uassertStatusOK(getStatusFromCommandResult(o)); } return o; } void DBClientCursor::peek(vector& v, int atMost) { invariant(_isInitialized); auto end = atMost >= static_cast(_batch.objs.size() - _batch.pos) ? _batch.objs.end() : _batch.objs.begin() + _batch.pos + atMost; v.insert(v.end(), _batch.objs.begin() + _batch.pos, end); } BSONObj DBClientCursor::peekFirst() { vector v; peek(v, 1); if (v.size() > 0) return v[0]; else return BSONObj(); } bool DBClientCursor::peekError(BSONObj* error) { invariant(_isInitialized); if (!_wasError) return false; vector v; peek(v, 1); verify(v.size() == 1); // We check both the legacy error format, and the new error format. hasErrField checks for // $err, and getStatusFromCommandResult checks for modern errors of the form '{ok: 0.0, code: // <...>, errmsg: ...}'. verify(hasErrField(v[0]) || !getStatusFromCommandResult(v[0]).isOK()); if (error) *error = v[0].getOwned(); return true; } void DBClientCursor::attach(AScopedConnection* conn) { verify(_scopedHost.size() == 0); verify(conn); verify(conn->get()); if (conn->get()->type() == ConnectionString::ConnectionType::kReplicaSet) { if (_client) _scopedHost = _client->getServerAddress(); else massert(14821, "No client specified, cannot store multi-host connection.", false); } else { _scopedHost = conn->getHost(); } conn->done(); _client = nullptr; } DBClientCursor::DBClientCursor(DBClientBase* client, const NamespaceStringOrUUID& nsOrUuid, long long cursorId, bool isExhaust, std::vector initialBatch, boost::optional operationTime, boost::optional postBatchResumeToken) : _batch{std::move(initialBatch)}, _client(client), _originalHost(_client->getServerAddress()), _nsOrUuid(nsOrUuid), _isInitialized(true), _ns(nsOrUuid.nss() ? *nsOrUuid.nss() : NamespaceString(nsOrUuid.dbName())), _cursorId(cursorId), _isExhaust(isExhaust), _operationTime(operationTime), _postBatchResumeToken(postBatchResumeToken) {} DBClientCursor::DBClientCursor(DBClientBase* client, FindCommandRequest findRequest, const ReadPreferenceSetting& readPref, bool isExhaust) : _client(client), _originalHost(_client->getServerAddress()), _nsOrUuid(findRequest.getNamespaceOrUUID()), _ns(_nsOrUuid.nss() ? *_nsOrUuid.nss() : NamespaceString(_nsOrUuid.dbName())), _batchSize(findRequest.getBatchSize().value_or(0)), _findRequest(std::move(findRequest)), _readPref(readPref), _isExhaust(isExhaust) { // Internal clients should always pass an explicit readConcern. If the caller did not already // pass a readConcern than we must explicitly initialize an empty readConcern so that it ends up // in the serialized version of the find command which will be sent across the wire. if (!_findRequest->getReadConcern()) { _findRequest->setReadConcern(BSONObj{}); } } StatusWith> DBClientCursor::fromAggregationRequest( DBClientBase* client, AggregateCommandRequest aggRequest, bool secondaryOk, bool useExhaust) { BSONObj ret; try { if (!client->runCommand(aggRequest.getNamespace().dbName(), aggregation_request_helper::serializeToCommandObj(aggRequest), ret, secondaryOk ? QueryOption_SecondaryOk : 0)) { return getStatusFromCommandResult(ret); } } catch (...) { return exceptionToStatus(); } long long cursorId = ret["cursor"].Obj()["id"].Long(); std::vector firstBatch; for (BSONElement elem : ret["cursor"].Obj()["firstBatch"].Array()) { firstBatch.emplace_back(elem.Obj().getOwned()); } boost::optional postBatchResumeToken; if (auto postBatchResumeTokenElem = ret["cursor"].Obj()["postBatchResumeToken"]; postBatchResumeTokenElem.type() == BSONType::Object) { postBatchResumeToken = postBatchResumeTokenElem.Obj().getOwned(); } else if (ret["cursor"].Obj().hasField("postBatchResumeToken")) { return Status(ErrorCodes::Error(5761702), "Expected field 'postbatchResumeToken' to be of object type"); } boost::optional operationTime = boost::none; if (ret.hasField(LogicalTime::kOperationTimeFieldName)) { operationTime = LogicalTime::fromOperationTime(ret).asTimestamp(); } return {std::make_unique(client, aggRequest.getNamespace(), cursorId, useExhaust, firstBatch, operationTime, postBatchResumeToken)}; } DBClientCursor::~DBClientCursor() { kill(); } void DBClientCursor::kill() { DESTRUCTOR_GUARD({ if (_cursorId && !globalInShutdownDeprecated()) { auto killCursor = [&](auto&& conn) { conn->killCursor(_ns, _cursorId); }; // We only need to kill the cursor if there aren't pending replies. Pending replies // indicates that this is an exhaust cursor, so the connection must be closed and the // cursor will automatically be cleaned up by the upstream server. if (_client && !_connectionHasPendingReplies) { killCursor(_client); } } }); // Mark this cursor as dead since we can't do any getMores. _cursorId = 0; _isInitialized = false; } } // namespace mongo