Edit File: fstrm.h
/* * Copyright (c) 2013-2014 by Farsight Security, Inc. * * Permission is hereby granted, free of charge, to any person obtaining * a copy of this software and associated documentation files (the * "Software"), to deal in the Software without restriction, including * without limitation the rights to use, copy, modify, merge, publish, * distribute, sublicense, and/or sell copies of the Software, and to * permit persons to whom the Software is furnished to do so, subject to * the following conditions: * * The above copyright notice and this permission notice shall be included * in all copies or substantial portions of the Software. * * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. * IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY * CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. * */ /*! \file * \mainpage Introduction * * This is `fstrm`, a C implementation of the Frame Streams data transport * protocol. * * Frame Streams is a light weight, binary clean protocol that allows for the * transport of arbitrarily encoded data payload sequences with minimal framing * overhead -- just four bytes per data frame. Frame Streams does not specify an * encoding format for data frames and can be used with any data serialization * format that produces byte sequences, such as [Protocol Buffers], [XML], * [JSON], [MessagePack], [YAML], etc. Frame Streams can be used as both a * streaming transport over a reliable byte stream socket (TCP sockets, TLS * connections, `AF_UNIX` sockets, etc.) for data in motion as well as a file * format for data at rest. A "Content Type" header identifies the type of * payload being carried over an individual Frame Stream and allows cooperating * programs to determine how to interpret a given sequence of data payloads. * * `fstrm` is an optimized C implementation of Frame Streams that includes a * fast, lockless circular queue implementation and exposes library interfaces * for setting up a dedicated Frame Streams I/O thread and asynchronously * submitting data frames for transport from worker threads. It was originally * written to facilitate the addition of high speed binary logging to DNS * servers written in C using the [dnstap] log format. * * This is the API documentation for the `fstrm` library. For the project * hosting site, see <https://github.com/farsightsec/fstrm>. * * \authors Farsight Security, Inc. and the `fstrm` authors. * * \copyright 2013-2018. Licensed under the terms of the [MIT] license. * * [Protocol Buffers]: https://developers.google.com/protocol-buffers/ * [XML]: http://www.w3.org/TR/xml11/ * [JSON]: http://www.json.org/ * [MessagePack]: http://msgpack.org/ * [YAML]: http://www.yaml.org/ * [dnstap]: http://dnstap.info/ * [MIT]: https://opensource.org/licenses/MIT * * \page overview Library overview * * \section init Initializing the library * * `fstrm` has no global library state. In most cases, only a single * \ref fstrm_iothr library context object will be needed for the entire process, * which will implicitly create a background I/O serialization thread. This I/O * thread is bound to a particular output writer (for example, an `AF_UNIX` * socket) and is fully buffered -- submitted data frames will be accumulated in * an output buffer and periodically flushed, minimizing the number of system * calls that need to be performed. This frees worker threads from waiting for a * write() to complete. * * `fstrm` abstracts the actual I/O operations needed to read or write a byte * stream. File and socket I/O implementations are included in the library, but * if necessary `fstrm` can be extended to support new types of byte stream * transports. See the \ref fstrm_reader, \ref fstrm_writer, and \ref fstrm_rdwr * interfaces for details. * * The following code example shows the initialization of an `fstrm_iothr` * library context object connected to an \ref fstrm_file writer. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ const char *file_path = "/tmp/output.fs"; struct fstrm_file_options *fopt; struct fstrm_iothr *iothr; struct fstrm_writer *writer; fopt = fstrm_file_options_init(); fstrm_file_options_set_file_path(fopt, file_path); writer = fstrm_file_writer_init(fopt, NULL); if (!writer) { fprintf(stderr, "Error: fstrm_file_writer_init() failed.\n"); exit(EXIT_FAILURE); } iothr = fstrm_iothr_init(NULL, &writer); if (!iothr) { fprintf(stderr, "Error: fstrm_iothr_init() failed.\n"); exit(EXIT_FAILURE); } fstrm_file_options_destroy(&fopt); ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ * * Since the I/O operations are abstracted through the `fstrm_writer` interface, * the `writer` variable in the above example could instead have been * initialized with a completely different implementation. For example, * \ref fstrm_unix_writer objects can be initialized as follows: ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ const char *socket_path = "/tmp/output.sock"; struct fstrm_writer *writer; struct fstrm_unix_writer_options *uwopt; uwopt = fstrm_unix_writer_options_init(); fstrm_unix_writer_options_set_socket_path(uwopt, socket_path); writer = fstrm_unix_writer_init(uwopt, NULL); if (!writer) { fprintf(stderr, "Error: fstrm_unix_writer_init() failed.\n"); exit(EXIT_FAILURE); } ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ * * \section queue Getting an input queue * * After the `fstrm_iothr` object has been created with fstrm_iothr_init(), an * input queue handle can be obtained with the fstrm_iothr_get_input_queue() * function, which returns an `fstrm_iothr_queue` object. This function is * thread-safe and returns a unique queue each time it is called, up to the * number of queues specified by fstrm_iothr_options_set_num_input_queues(). * `fstrm_iothr_queue` objects belong to their parent `fstrm_iothr` object and * will be destroyed when the parent `fstrm_iothr` object is destroyed. * * The following code example shows a single `fstrm_iothr_queue` handle being * obtained from an already initialized `fstrm_iothr` library context object. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ // 'iothr' is a struct fstrm_iothr * struct fstrm_iothr_queue *ioq; ioq = fstrm_iothr_get_input_queue(iothr); if (!ioq) { fprintf(stderr, "Error: fstrm_iothr_get_input_queue() failed.\n"); exit(EXIT_FAILURE); } ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ * * \section submit Submitting data frames * * Once the `fstrm_iothr` object has been created and an `fstrm_iothr_queue` * handle is available, data frames can be submitted for asynchronous writing * using the fstrm_iothr_submit() function. A callback is passed to this * function which will be invoked to deallocate the data frame once the I/O * thread has completed processing it. In the common case where the data frame * is dynamically allocated with `malloc()`, the deallocation callback must call * `free()`. fstrm_free_wrapper() is provided as a convenience function which * does this and can be specified as the `free_func` parameter to * fstrm_iothr_submit(). * * If space is available in the queue, fstrm_iothr_submit() will return * #fstrm_res_success, indicating that ownership of the memory allocation for the * data frame has passed from the caller to the library. The caller must not * reuse or deallocate the memory for the data frame after a successful call to * fstrm_iothr_submit(). * * Callers must check the return value of fstrm_iothr_submit(). If this function * fails, that is, it returns any result code other than #fstrm_res_success, the * caller must deallocate or otherwise dispose of memory allocated for the data * frame, in order to avoid leaking memory. fstrm_iothr_submit() can fail with * #fstrm_res_again if there is currently no space in the circular queue for an * additional frame, in which case a later call to fstrm_iothr_submit() with the * same parameters may succeed. However, if fstrm_iothr_submit() fails with * #fstrm_res_invalid, then there is a problem with the parameters and a later * call will not succeed. * * The following code example shows data frames containing a short sequence of * bytes being created and submitted repeatedly, with appropriate error * handling. Note that the data frames in this example intentionally contain * embedded unprintable characters, showing that Frame Streams is binary clean. * This example follows from the previous examples, where the `iothr` and `ioq` * variables have already been initialized. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ // 'iothr' is a struct fstrm_iothr * // 'ioq' is a struct fstrm_queue * const unsigned num_frames = 100; const uint8_t frame_template[] = { 'H', 'e', 'l', 'l', 'o', 0x00, 0x01, 0x02, 0x03, 'W', 'o', 'r', 'l', 'd', 0x04, 0x05, 0x06, 0x07, }; for (unsigned i = 0; i < num_frames; i++) { // Allocate a new frame from the template. uint8_t *frame = malloc(sizeof(frame_template)); if (!frame) break; memcpy(frame, frame_template, sizeof(frame_template)); // Submit the frame for writing. for (;;) { fstrm_res res; res = fstrm_iothr_submit(iothr, ioq, frame, sizeof(frame_template), fstrm_free_wrapper, NULL); if (res == fstrm_res_success) { // Frame successfully queued. break; } else if (res == fstrm_res_again) { // Queue is full. Try again in a busy loop. // Alternatively, if loss can be tolerated we // could free the frame here and break out of // the loop. continue; } else { // Permanent failure. free(frame); fputs("fstrm_iothr_submit() failed.\n", stderr); break; } } } ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ * * \section shutdown Shutting down * * Calling fstrm_iothr_destroy() on the `fstrm_iothr` object will signal the I/O * thread to flush any outstanding data frames being written and will deallocate * all associated resources. This function is synchronous and does not return * until the I/O thread has terminated. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ // 'iothr' is a struct fstrm_iothr * fstrm_iothr_destroy(&iothr); ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ */ #ifndef FSTRM_H #define FSTRM_H #ifdef __cplusplus extern "C" { #endif #include <sys/uio.h> #include <stddef.h> #include <stdint.h> /** * \defgroup fstrm_res fstrm_res * * Library result codes. * @{ */ /** * Result codes for functions. */ typedef enum { /** Success. */ fstrm_res_success, /** Failure. */ fstrm_res_failure, /** Resource temporarily unavailable. */ fstrm_res_again, /** Parameters were invalid. */ fstrm_res_invalid, /** The end of a stream has been reached. */ fstrm_res_stop, } fstrm_res; /**@}*/ struct fstrm_control; struct fstrm_file_options; struct fstrm_iothr; struct fstrm_iothr_options; struct fstrm_iothr_queue; struct fstrm_rdwr; struct fstrm_reader_options; struct fstrm_unix_writer_options; struct fstrm_writer; struct fstrm_writer_options; #include <fstrm/control.h> #include <fstrm/file.h> #include <fstrm/iothr.h> #include <fstrm/rdwr.h> #include <fstrm/reader.h> #include <fstrm/tcp_writer.h> #include <fstrm/unix_writer.h> #include <fstrm/writer.h> #ifdef __cplusplus } #endif #endif /* FSTRM_H */