#include "HTTPHandler.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include namespace DB { namespace ErrorCodes { extern const int LOGICAL_ERROR; extern const int CANNOT_PARSE_TEXT; extern const int CANNOT_PARSE_ESCAPE_SEQUENCE; extern const int CANNOT_PARSE_QUOTED_STRING; extern const int CANNOT_PARSE_DATE; extern const int CANNOT_PARSE_DATETIME; extern const int CANNOT_PARSE_NUMBER; extern const int CANNOT_OPEN_FILE; extern const int UNKNOWN_ELEMENT_IN_AST; extern const int UNKNOWN_TYPE_OF_AST_NODE; extern const int TOO_DEEP_AST; extern const int TOO_BIG_AST; extern const int UNEXPECTED_AST_STRUCTURE; extern const int SYNTAX_ERROR; extern const int INCORRECT_DATA; extern const int TYPE_MISMATCH; extern const int UNKNOWN_TABLE; extern const int UNKNOWN_FUNCTION; extern const int UNKNOWN_IDENTIFIER; extern const int UNKNOWN_TYPE; extern const int UNKNOWN_STORAGE; extern const int UNKNOWN_DATABASE; extern const int UNKNOWN_SETTING; extern const int UNKNOWN_DIRECTION_OF_SORTING; extern const int UNKNOWN_AGGREGATE_FUNCTION; extern const int UNKNOWN_FORMAT; extern const int UNKNOWN_DATABASE_ENGINE; extern const int UNKNOWN_TYPE_OF_QUERY; extern const int QUERY_IS_TOO_LARGE; extern const int NOT_IMPLEMENTED; extern const int SOCKET_TIMEOUT; extern const int UNKNOWN_USER; extern const int WRONG_PASSWORD; extern const int REQUIRED_PASSWORD; extern const int INVALID_SESSION_TIMEOUT; extern const int HTTP_LENGTH_REQUIRED; } static Poco::Net::HTTPResponse::HTTPStatus exceptionCodeToHTTPStatus(int exception_code) { using namespace Poco::Net; if (exception_code == ErrorCodes::REQUIRED_PASSWORD) return HTTPResponse::HTTP_UNAUTHORIZED; else if (exception_code == ErrorCodes::CANNOT_PARSE_TEXT || exception_code == ErrorCodes::CANNOT_PARSE_ESCAPE_SEQUENCE || exception_code == ErrorCodes::CANNOT_PARSE_QUOTED_STRING || exception_code == ErrorCodes::CANNOT_PARSE_DATE || exception_code == ErrorCodes::CANNOT_PARSE_DATETIME || exception_code == ErrorCodes::CANNOT_PARSE_NUMBER || exception_code == ErrorCodes::UNKNOWN_ELEMENT_IN_AST || exception_code == ErrorCodes::UNKNOWN_TYPE_OF_AST_NODE || exception_code == ErrorCodes::TOO_DEEP_AST || exception_code == ErrorCodes::TOO_BIG_AST || exception_code == ErrorCodes::UNEXPECTED_AST_STRUCTURE || exception_code == ErrorCodes::SYNTAX_ERROR || exception_code == ErrorCodes::INCORRECT_DATA || exception_code == ErrorCodes::TYPE_MISMATCH) return HTTPResponse::HTTP_BAD_REQUEST; else if (exception_code == ErrorCodes::UNKNOWN_TABLE || exception_code == ErrorCodes::UNKNOWN_FUNCTION || exception_code == ErrorCodes::UNKNOWN_IDENTIFIER || exception_code == ErrorCodes::UNKNOWN_TYPE || exception_code == ErrorCodes::UNKNOWN_STORAGE || exception_code == ErrorCodes::UNKNOWN_DATABASE || exception_code == ErrorCodes::UNKNOWN_SETTING || exception_code == ErrorCodes::UNKNOWN_DIRECTION_OF_SORTING || exception_code == ErrorCodes::UNKNOWN_AGGREGATE_FUNCTION || exception_code == ErrorCodes::UNKNOWN_FORMAT || exception_code == ErrorCodes::UNKNOWN_DATABASE_ENGINE || exception_code == ErrorCodes::UNKNOWN_TYPE_OF_QUERY) return HTTPResponse::HTTP_NOT_FOUND; else if (exception_code == ErrorCodes::QUERY_IS_TOO_LARGE) return HTTPResponse::HTTP_REQUESTENTITYTOOLARGE; else if (exception_code == ErrorCodes::NOT_IMPLEMENTED) return HTTPResponse::HTTP_NOT_IMPLEMENTED; else if (exception_code == ErrorCodes::SOCKET_TIMEOUT || exception_code == ErrorCodes::CANNOT_OPEN_FILE) return HTTPResponse::HTTP_SERVICE_UNAVAILABLE; else if (exception_code == ErrorCodes::HTTP_LENGTH_REQUIRED) return HTTPResponse::HTTP_LENGTH_REQUIRED; return HTTPResponse::HTTP_INTERNAL_SERVER_ERROR; } static std::chrono::steady_clock::duration parseSessionTimeout( const Poco::Util::AbstractConfiguration & config, const HTMLForm & params) { unsigned session_timeout = config.getInt("default_session_timeout", 60); if (params.has("session_timeout")) { unsigned max_session_timeout = config.getUInt("max_session_timeout", 3600); std::string session_timeout_str = params.get("session_timeout"); ReadBufferFromString buf(session_timeout_str); if (!tryReadIntText(session_timeout, buf) || !buf.eof()) throw Exception("Invalid session timeout: '" + session_timeout_str + "'", ErrorCodes::INVALID_SESSION_TIMEOUT); if (session_timeout > max_session_timeout) throw Exception("Session timeout '" + session_timeout_str + "' is larger than max_session_timeout: " + toString(max_session_timeout) + ". Maximum session timeout could be modified in configuration file.", ErrorCodes::INVALID_SESSION_TIMEOUT); } return std::chrono::seconds(session_timeout); } void HTTPHandler::pushDelayedResults(Output & used_output) { std::vector write_buffers; std::vector read_buffers; std::vector read_buffers_raw_ptr; auto cascade_buffer = typeid_cast(used_output.out_maybe_delayed_and_compressed.get()); if (!cascade_buffer) throw Exception("Expected CascadeWriteBuffer", ErrorCodes::LOGICAL_ERROR); cascade_buffer->getResultBuffers(write_buffers); if (write_buffers.empty()) throw Exception("At least one buffer is expected to overwrite result into HTTP response", ErrorCodes::LOGICAL_ERROR); for (auto & write_buf : write_buffers) { IReadableWriteBuffer * write_buf_concrete; ReadBufferPtr reread_buf; if (write_buf && (write_buf_concrete = dynamic_cast(write_buf.get())) && (reread_buf = write_buf_concrete->tryGetReadBuffer())) { read_buffers.emplace_back(reread_buf); read_buffers_raw_ptr.emplace_back(reread_buf.get()); } } ConcatReadBuffer concat_read_buffer(read_buffers_raw_ptr); copyData(concat_read_buffer, *used_output.out_maybe_compressed); } HTTPHandler::HTTPHandler(IServer & server_) : server(server_) , log(&Logger::get("HTTPHandler")) { server_display_name = server.config().getString("display_name", getFQDNOrHostName()); } void HTTPHandler::processQuery( Poco::Net::HTTPServerRequest & request, HTMLForm & params, Poco::Net::HTTPServerResponse & response, Output & used_output) { Context context = server.context(); CurrentThread::QueryScope query_scope(context); LOG_TRACE(log, "Request URI: " << request.getURI()); std::istream & istr = request.stream(); /// Part of the query can be passed in the 'query' parameter and the rest in the request body /// (http method need not necessarily be POST). In this case the entire query consists of the /// contents of the 'query' parameter, a line break and the request body. std::string query_param = params.get("query", ""); if (!query_param.empty()) query_param += '\n'; /// The user and password can be passed by headers (similar to X-Auth-*), /// which is used by load balancers to pass authentication information. std::string user = request.get("X-ClickHouse-User", ""); std::string password = request.get("X-ClickHouse-Key", ""); std::string quota_key = request.get("X-ClickHouse-Quota", ""); if (user.empty() && password.empty() && quota_key.empty()) { /// User name and password can be passed using query parameters /// or using HTTP Basic auth (both methods are insecure). if (request.hasCredentials()) { Poco::Net::HTTPBasicCredentials credentials(request); user = credentials.getUsername(); password = credentials.getPassword(); } else { user = params.get("user", "default"); password = params.get("password", ""); } quota_key = params.get("quota_key", ""); } else { /// It is prohibited to mix different authorization schemes. if (request.hasCredentials() || params.has("user") || params.has("password") || params.has("quota_key")) { throw Exception("Invalid authentication: it is not allowed to use X-ClickHouse HTTP headers and other authentication methods simultaneously", ErrorCodes::REQUIRED_PASSWORD); } } std::string query_id = params.get("query_id", ""); context.setUser(user, password, request.clientAddress(), quota_key); context.setCurrentQueryId(query_id); /// The user could specify session identifier and session timeout. /// It allows to modify settings, create temporary tables and reuse them in subsequent requests. std::shared_ptr session; String session_id; std::chrono::steady_clock::duration session_timeout; bool session_is_set = params.has("session_id"); const auto & config = server.config(); if (session_is_set) { session_id = params.get("session_id"); session_timeout = parseSessionTimeout(config, params); std::string session_check = params.get("session_check", ""); session = context.acquireNamedSession(session_id, session_timeout, session_check == "1"); context = session->context; context.setSessionContext(session->context); } SCOPE_EXIT({ if (session) session->release(); }); /// The client can pass a HTTP header indicating supported compression method (gzip or deflate). String http_response_compression_methods = request.get("Accept-Encoding", ""); CompressionMethod http_response_compression_method = CompressionMethod::None; if (!http_response_compression_methods.empty()) { /// If client supports brotli - it's preferred. /// Both gzip and deflate are supported. If the client supports both, gzip is preferred. /// NOTE parsing of the list of methods is slightly incorrect. if (std::string::npos != http_response_compression_methods.find("br")) http_response_compression_method = CompressionMethod::Brotli; else if (std::string::npos != http_response_compression_methods.find("gzip")) http_response_compression_method = CompressionMethod::Gzip; else if (std::string::npos != http_response_compression_methods.find("deflate")) http_response_compression_method = CompressionMethod::Zlib; } bool client_supports_http_compression = http_response_compression_method != CompressionMethod::None; /// Client can pass a 'compress' flag in the query string. In this case the query result is /// compressed using internal algorithm. This is not reflected in HTTP headers. bool internal_compression = params.getParsed("compress", false); /// At least, we should postpone sending of first buffer_size result bytes size_t buffer_size_total = std::max( params.getParsed("buffer_size", DBMS_DEFAULT_BUFFER_SIZE), static_cast(DBMS_DEFAULT_BUFFER_SIZE)); /// If it is specified, the whole result will be buffered. /// First ~buffer_size bytes will be buffered in memory, the remaining bytes will be stored in temporary file. bool buffer_until_eof = params.getParsed("wait_end_of_query", false); size_t buffer_size_http = DBMS_DEFAULT_BUFFER_SIZE; size_t buffer_size_memory = (buffer_size_total > buffer_size_http) ? buffer_size_total : 0; unsigned keep_alive_timeout = config.getUInt("keep_alive_timeout", 10); used_output.out = std::make_shared( request, response, keep_alive_timeout, client_supports_http_compression, http_response_compression_method); if (internal_compression) used_output.out_maybe_compressed = std::make_shared(*used_output.out); else used_output.out_maybe_compressed = used_output.out; if (buffer_size_memory > 0 || buffer_until_eof) { CascadeWriteBuffer::WriteBufferPtrs cascade_buffer1; CascadeWriteBuffer::WriteBufferConstructors cascade_buffer2; if (buffer_size_memory > 0) cascade_buffer1.emplace_back(std::make_shared(buffer_size_memory)); if (buffer_until_eof) { const std::string tmp_path(context.getTemporaryVolume()->getNextDisk()->getPath()); const std::string tmp_path_template(tmp_path + "http_buffers/"); auto create_tmp_disk_buffer = [tmp_path_template] (const WriteBufferPtr &) { return WriteBufferFromTemporaryFile::create(tmp_path_template); }; cascade_buffer2.emplace_back(std::move(create_tmp_disk_buffer)); } else { auto push_memory_buffer_and_continue = [next_buffer = used_output.out_maybe_compressed] (const WriteBufferPtr & prev_buf) { auto prev_memory_buffer = typeid_cast(prev_buf.get()); if (!prev_memory_buffer) throw Exception("Expected MemoryWriteBuffer", ErrorCodes::LOGICAL_ERROR); auto rdbuf = prev_memory_buffer->tryGetReadBuffer(); copyData(*rdbuf , *next_buffer); return next_buffer; }; cascade_buffer2.emplace_back(push_memory_buffer_and_continue); } used_output.out_maybe_delayed_and_compressed = std::make_shared( std::move(cascade_buffer1), std::move(cascade_buffer2)); } else { used_output.out_maybe_delayed_and_compressed = used_output.out_maybe_compressed; } std::unique_ptr in_param = std::make_unique(query_param); std::unique_ptr in_post_raw = std::make_unique(istr); /// Request body can be compressed using algorithm specified in the Content-Encoding header. String http_request_compression_method_str = request.get("Content-Encoding", ""); std::unique_ptr in_post = wrapReadBufferWithCompressionMethod( std::make_unique(istr), chooseCompressionMethod({}, http_request_compression_method_str)); /// The data can also be compressed using incompatible internal algorithm. This is indicated by /// 'decompress' query parameter. std::unique_ptr in_post_maybe_compressed; bool in_post_compressed = false; if (params.getParsed("decompress", false)) { in_post_maybe_compressed = std::make_unique(*in_post); in_post_compressed = true; } else in_post_maybe_compressed = std::move(in_post); std::unique_ptr in; static const NameSet reserved_param_names{"query", "compress", "decompress", "user", "password", "quota_key", "query_id", "stacktrace", "buffer_size", "wait_end_of_query", "session_id", "session_timeout", "session_check"}; Names reserved_param_suffixes; auto param_could_be_skipped = [&] (const String & name) { if (reserved_param_names.count(name)) return true; for (const String & suffix : reserved_param_suffixes) { if (endsWith(name, suffix)) return true; } return false; }; /// Settings can be overridden in the query. /// Some parameters (database, default_format, everything used in the code above) do not /// belong to the Settings class. /// 'readonly' setting values mean: /// readonly = 0 - any query is allowed, client can change any setting. /// readonly = 1 - only readonly queries are allowed, client can't change settings. /// readonly = 2 - only readonly queries are allowed, client can change any setting except 'readonly'. /// In theory if initially readonly = 0, the client can change any setting and then set readonly /// to some other value. auto & settings = context.getSettingsRef(); /// Only readonly queries are allowed for HTTP GET requests. if (request.getMethod() == Poco::Net::HTTPServerRequest::HTTP_GET) { if (settings.readonly == 0) settings.readonly = 2; } bool has_external_data = startsWith(request.getContentType().data(), "multipart/form-data"); if (has_external_data) { /// Skip unneeded parameters to avoid confusing them later with context settings or query parameters. reserved_param_suffixes.reserve(3); /// It is a bug and ambiguity with `date_time_input_format` and `low_cardinality_allow_in_native_format` formats/settings. reserved_param_suffixes.emplace_back("_format"); reserved_param_suffixes.emplace_back("_types"); reserved_param_suffixes.emplace_back("_structure"); } SettingsChanges settings_changes; for (const auto & [key, value] : params) { if (key == "database") { context.setCurrentDatabase(value); } else if (key == "default_format") { context.setDefaultFormat(value); } else if (param_could_be_skipped(key)) { } else if (startsWith(key, "param_")) { /// Save name and values of substitution in dictionary. const String parameter_name = key.substr(strlen("param_")); context.setQueryParameter(parameter_name, value); } else { /// All other query parameters are treated as settings. settings_changes.push_back({key, value}); } } /// For external data we also want settings context.checkSettingsConstraints(settings_changes); context.applySettingsChanges(settings_changes); /// Used in case of POST request with form-data, but it isn't expected to be deleted after that scope. std::string full_query; /// Support for "external data for query processing". if (has_external_data) { ExternalTablesHandler handler(context, params); params.load(request, istr, handler); /// Params are of both form params POST and uri (GET params) for (const auto & it : params) if (it.first == "query") full_query += it.second; in = std::make_unique(full_query); } else in = std::make_unique(*in_param, *in_post_maybe_compressed); /// HTTP response compression is turned on only if the client signalled that they support it /// (using Accept-Encoding header) and 'enable_http_compression' setting is turned on. used_output.out->setCompression(client_supports_http_compression && settings.enable_http_compression); if (client_supports_http_compression) used_output.out->setCompressionLevel(settings.http_zlib_compression_level); used_output.out->setSendProgressInterval(settings.http_headers_progress_interval_ms); /// If 'http_native_compression_disable_checksumming_on_decompress' setting is turned on, /// checksums of client data compressed with internal algorithm are not checked. if (in_post_compressed && settings.http_native_compression_disable_checksumming_on_decompress) static_cast(*in_post_maybe_compressed).disableChecksumming(); /// Add CORS header if 'add_http_cors_header' setting is turned on and the client passed /// Origin header. used_output.out->addHeaderCORS(settings.add_http_cors_header && !request.get("Origin", "").empty()); ClientInfo & client_info = context.getClientInfo(); client_info.query_kind = ClientInfo::QueryKind::INITIAL_QUERY; client_info.interface = ClientInfo::Interface::HTTP; /// Query sent through HTTP interface is initial. client_info.initial_user = client_info.current_user; client_info.initial_query_id = client_info.current_query_id; client_info.initial_address = client_info.current_address; ClientInfo::HTTPMethod http_method = ClientInfo::HTTPMethod::UNKNOWN; if (request.getMethod() == Poco::Net::HTTPServerRequest::HTTP_GET) http_method = ClientInfo::HTTPMethod::GET; else if (request.getMethod() == Poco::Net::HTTPServerRequest::HTTP_POST) http_method = ClientInfo::HTTPMethod::POST; client_info.http_method = http_method; client_info.http_user_agent = request.get("User-Agent", ""); auto appendCallback = [&context] (ProgressCallback callback) { auto prev = context.getProgressCallback(); context.setProgressCallback([prev, callback] (const Progress & progress) { if (prev) prev(progress); callback(progress); }); }; /// While still no data has been sent, we will report about query execution progress by sending HTTP headers. if (settings.send_progress_in_http_headers) appendCallback([&used_output] (const Progress & progress) { used_output.out->onProgress(progress); }); if (settings.readonly > 0 && settings.cancel_http_readonly_queries_on_client_close) { Poco::Net::StreamSocket & socket = dynamic_cast(request).socket(); appendCallback([&context, &socket](const Progress &) { /// Assume that at the point this method is called no one is reading data from the socket any more. /// True for read-only queries. try { char b; //FIXME looks like MSG_DONTWAIT is useless because of POCO_BROKEN_TIMEOUTS int status = socket.receiveBytes(&b, 1, MSG_DONTWAIT | MSG_PEEK); if (status == 0) context.killCurrentQuery(); } catch (Poco::TimeoutException &) { } catch (...) { context.killCurrentQuery(); } }); } customizeContext(context); executeQuery(*in, *used_output.out_maybe_delayed_and_compressed, /* allow_into_outfile = */ false, context, [&response] (const String & current_query_id, const String & content_type, const String & format, const String & timezone) { response.setContentType(content_type); response.add("X-ClickHouse-Query-Id", current_query_id); response.add("X-ClickHouse-Format", format); response.add("X-ClickHouse-Timezone", timezone); } ); if (used_output.hasDelayed()) { /// TODO: set Content-Length if possible pushDelayedResults(used_output); } /// Send HTTP headers with code 200 if no exception happened and the data is still not sent to /// the client. used_output.out->finalize(); } void HTTPHandler::trySendExceptionToClient(const std::string & s, int exception_code, Poco::Net::HTTPServerRequest & request, Poco::Net::HTTPServerResponse & response, Output & used_output) { try { response.set("X-ClickHouse-Exception-Code", toString(exception_code)); /// If HTTP method is POST and Keep-Alive is turned on, we should read the whole request body /// to avoid reading part of the current request body in the next request. if (request.getMethod() == Poco::Net::HTTPRequest::HTTP_POST && response.getKeepAlive() && !request.stream().eof() && exception_code != ErrorCodes::HTTP_LENGTH_REQUIRED) { request.stream().ignore(std::numeric_limits::max()); } bool auth_fail = exception_code == ErrorCodes::UNKNOWN_USER || exception_code == ErrorCodes::WRONG_PASSWORD || exception_code == ErrorCodes::REQUIRED_PASSWORD; if (auth_fail) { response.requireAuthentication("ClickHouse server HTTP API"); } else { response.setStatusAndReason(exceptionCodeToHTTPStatus(exception_code)); } if (!response.sent() && !used_output.out_maybe_compressed) { /// If nothing was sent yet and we don't even know if we must compress the response. response.send() << s << std::endl; } else if (used_output.out_maybe_compressed) { /// Destroy CascadeBuffer to actualize buffers' positions and reset extra references if (used_output.hasDelayed()) used_output.out_maybe_delayed_and_compressed.reset(); /// Send the error message into already used (and possibly compressed) stream. /// Note that the error message will possibly be sent after some data. /// Also HTTP code 200 could have already been sent. /// If buffer has data, and that data wasn't sent yet, then no need to send that data bool data_sent = used_output.out->count() != used_output.out->offset(); if (!data_sent) { used_output.out_maybe_compressed->position() = used_output.out_maybe_compressed->buffer().begin(); used_output.out->position() = used_output.out->buffer().begin(); } writeString(s, *used_output.out_maybe_compressed); writeChar('\n', *used_output.out_maybe_compressed); used_output.out_maybe_compressed->next(); used_output.out->next(); used_output.out->finalize(); } } catch (...) { tryLogCurrentException(log, "Cannot send exception to client"); } } void HTTPHandler::handleRequest(Poco::Net::HTTPServerRequest & request, Poco::Net::HTTPServerResponse & response) { setThreadName("HTTPHandler"); ThreadStatus thread_status; Output used_output; /// In case of exception, send stack trace to client. bool with_stacktrace = false; try { response.setContentType("text/plain; charset=UTF-8"); response.set("X-ClickHouse-Server-Display-Name", server_display_name); /// For keep-alive to work. if (request.getVersion() == Poco::Net::HTTPServerRequest::HTTP_1_1) response.setChunkedTransferEncoding(true); HTMLForm params(request); with_stacktrace = params.getParsed("stacktrace", false); /// Workaround. Poco does not detect 411 Length Required case. if (request.getMethod() == Poco::Net::HTTPRequest::HTTP_POST && !request.getChunkedTransferEncoding() && !request.hasContentLength()) { throw Exception("The Transfer-Encoding is not chunked and there is no Content-Length header for POST request", ErrorCodes::HTTP_LENGTH_REQUIRED); } processQuery(request, params, response, used_output); LOG_INFO(log, "Done processing query"); } catch (...) { tryLogCurrentException(log); /** If exception is received from remote server, then stack trace is embedded in message. * If exception is thrown on local server, then stack trace is in separate field. */ std::string exception_message = getCurrentExceptionMessage(with_stacktrace, true); int exception_code = getCurrentExceptionCode(); trySendExceptionToClient(exception_message, exception_code, request, response, used_output); } } }