1// SPDX-License-Identifier: 0BSD 2 3/////////////////////////////////////////////////////////////////////////////// 4// 5/// \file common.h 6/// \brief Definitions common to the whole liblzma library 7// 8// Author: Lasse Collin 9// 10/////////////////////////////////////////////////////////////////////////////// 11 12#ifndef LZMA_COMMON_H 13#define LZMA_COMMON_H 14 15#include "sysdefs.h" 16#include "mythread.h" 17#include "tuklib_integer.h" 18 19// LZMA_API_EXPORT is used to mark the exported API functions. 20// It's used to define the LZMA_API macro. 21// 22// lzma_attr_visibility_hidden is used for marking *declarations* of extern 23// variables that are internal to liblzma (-fvisibility=hidden alone is 24// enough to hide the *definitions*). Such markings allow slightly more 25// efficient code to accesses those variables in ELF shared libraries. 26#if defined(_WIN32) || defined(__CYGWIN__) 27# ifdef DLL_EXPORT 28# define LZMA_API_EXPORT __declspec(dllexport) 29# else 30# define LZMA_API_EXPORT 31# endif 32# define lzma_attr_visibility_hidden 33// Don't use ifdef or defined() below. 34#elif HAVE_VISIBILITY 35# define LZMA_API_EXPORT __attribute__((__visibility__("default"))) 36# define lzma_attr_visibility_hidden \ 37 __attribute__((__visibility__("hidden"))) 38#else 39# define LZMA_API_EXPORT 40# define lzma_attr_visibility_hidden 41#endif 42 43#define LZMA_API(type) LZMA_API_EXPORT type LZMA_API_CALL 44 45#include "lzma.h" 46 47// This is for detecting modern GCC and Clang attributes 48// like __symver__ in GCC >= 10. 49#ifdef __has_attribute 50# define lzma_has_attribute(attr) __has_attribute(attr) 51#else 52# define lzma_has_attribute(attr) 0 53#endif 54 55// The extra symbol versioning in the C files may only be used when 56// building a shared library. If HAVE_SYMBOL_VERSIONS_LINUX is defined 57// to 2 then symbol versioning is done only if also PIC is defined. 58// By default Libtool defines PIC when building a shared library and 59// doesn't define it when building a static library but it can be 60// overridden with --with-pic and --without-pic. configure let's rely 61// on PIC if neither --with-pic or --without-pic was used. 62#if defined(HAVE_SYMBOL_VERSIONS_LINUX) \ 63 && (HAVE_SYMBOL_VERSIONS_LINUX == 2 && !defined(PIC)) 64# undef HAVE_SYMBOL_VERSIONS_LINUX 65#endif 66 67#ifdef HAVE_SYMBOL_VERSIONS_LINUX 68// To keep link-time optimization (LTO, -flto) working with GCC, 69// the __symver__ attribute must be used instead of __asm__(".symver ..."). 70// Otherwise the symbol versions may be lost, resulting in broken liblzma 71// that has wrong default versions in the exported symbol list! 72// The attribute was added in GCC 10; LTO with older GCC is not supported. 73// 74// To keep -Wmissing-prototypes happy, use LZMA_SYMVER_API only with function 75// declarations (including those with __alias__ attribute) and LZMA_API with 76// the function definitions. This means a little bit of silly copy-and-paste 77// between declarations and definitions though. 78// 79// As of GCC 12.2, the __symver__ attribute supports only @ and @@ but the 80// very convenient @@@ isn't supported (it's supported by GNU assembler 81// since 2000). When using @@ instead of @@@, the internal name must not be 82// the same as the external name to avoid problems in some situations. This 83// is why "#define foo_52 foo" is needed for the default symbol versions. 84// 85// __has_attribute is supported before GCC 10 and it is supported in Clang 14 86// too (which doesn't support __symver__) so use it to detect if __symver__ 87// is available. This should be far more reliable than looking at compiler 88// version macros as nowadays especially __GNUC__ is defined by many compilers. 89# if lzma_has_attribute(__symver__) 90# define LZMA_SYMVER_API(extnamever, type, intname) \ 91 extern __attribute__((__symver__(extnamever))) \ 92 LZMA_API(type) intname 93# else 94# define LZMA_SYMVER_API(extnamever, type, intname) \ 95 __asm__(".symver " #intname "," extnamever); \ 96 extern LZMA_API(type) intname 97# endif 98#endif 99 100// MSVC has __forceinline which shouldn't be combined with the inline keyword 101// (results in a warning). 102// 103// GCC 3.1 added always_inline attribute so we don't need to check 104// for __GNUC__ version. Similarly, all relevant Clang versions 105// support it (at least Clang 3.0.0 does already). 106// Other compilers might support too which also support __has_attribute 107// (Solaris Studio) so do that check too. 108#if defined(_MSC_VER) 109# define lzma_always_inline __forceinline 110#elif defined(__GNUC__) || defined(__clang__) || defined(__INTEL_COMPILER) \ 111 || lzma_has_attribute(__always_inline__) 112# define lzma_always_inline inline __attribute__((__always_inline__)) 113#else 114# define lzma_always_inline inline 115#endif 116 117// These allow helping the compiler in some often-executed branches, whose 118// result is almost always the same. 119#ifdef __GNUC__ 120# define likely(expr) __builtin_expect(expr, true) 121# define unlikely(expr) __builtin_expect(expr, false) 122#else 123# define likely(expr) (expr) 124# define unlikely(expr) (expr) 125#endif 126 127 128/// Size of temporary buffers needed in some filters 129#define LZMA_BUFFER_SIZE 4096 130 131 132/// Maximum number of worker threads within one multithreaded component. 133/// The limit exists solely to make it simpler to prevent integer overflows 134/// when allocating structures etc. This should be big enough for now... 135/// the code won't scale anywhere close to this number anyway. 136#define LZMA_THREADS_MAX 16384 137 138 139/// Starting value for memory usage estimates. Instead of calculating size 140/// of _every_ structure and taking into account malloc() overhead etc., we 141/// add a base size to all memory usage estimates. It's not very accurate 142/// but should be easily good enough. 143#define LZMA_MEMUSAGE_BASE (UINT64_C(1) << 15) 144 145/// Start of internal Filter ID space. These IDs must never be used 146/// in Streams. 147#define LZMA_FILTER_RESERVED_START (LZMA_VLI_C(1) << 62) 148 149 150/// Supported flags that can be passed to lzma_stream_decoder(), 151/// lzma_auto_decoder(), or lzma_stream_decoder_mt(). 152#define LZMA_SUPPORTED_FLAGS \ 153 ( LZMA_TELL_NO_CHECK \ 154 | LZMA_TELL_UNSUPPORTED_CHECK \ 155 | LZMA_TELL_ANY_CHECK \ 156 | LZMA_IGNORE_CHECK \ 157 | LZMA_CONCATENATED \ 158 | LZMA_FAIL_FAST ) 159 160 161/// Largest valid lzma_action value as unsigned integer. 162#define LZMA_ACTION_MAX ((unsigned int)(LZMA_FULL_BARRIER)) 163 164 165/// Special return value (lzma_ret) to indicate that a timeout was reached 166/// and lzma_code() must not return LZMA_BUF_ERROR. This is converted to 167/// LZMA_OK in lzma_code(). 168#define LZMA_TIMED_OUT LZMA_RET_INTERNAL1 169 170/// Special return value (lzma_ret) for use in stream_decoder_mt.c to 171/// indicate Index was detected instead of a Block Header. 172#define LZMA_INDEX_DETECTED LZMA_RET_INTERNAL2 173 174 175typedef struct lzma_next_coder_s lzma_next_coder; 176 177typedef struct lzma_filter_info_s lzma_filter_info; 178 179 180/// Type of a function used to initialize a filter encoder or decoder 181typedef lzma_ret (*lzma_init_function)( 182 lzma_next_coder *next, const lzma_allocator *allocator, 183 const lzma_filter_info *filters); 184 185/// Type of a function to do some kind of coding work (filters, Stream, 186/// Block encoders/decoders etc.). Some special coders use don't use both 187/// input and output buffers, but for simplicity they still use this same 188/// function prototype. 189typedef lzma_ret (*lzma_code_function)( 190 void *coder, const lzma_allocator *allocator, 191 const uint8_t *restrict in, size_t *restrict in_pos, 192 size_t in_size, uint8_t *restrict out, 193 size_t *restrict out_pos, size_t out_size, 194 lzma_action action); 195 196/// Type of a function to free the memory allocated for the coder 197typedef void (*lzma_end_function)( 198 void *coder, const lzma_allocator *allocator); 199 200 201/// Raw coder validates and converts an array of lzma_filter structures to 202/// an array of lzma_filter_info structures. This array is used with 203/// lzma_next_filter_init to initialize the filter chain. 204struct lzma_filter_info_s { 205 /// Filter ID. This can be used to share the same initiazation 206 /// function *and* data structures with different Filter IDs 207 /// (LZMA_FILTER_LZMA1EXT does it), and also by the encoder 208 /// with lzma_filters_update() if filter chain is updated 209 /// in the middle of a raw stream or Block (LZMA_SYNC_FLUSH). 210 lzma_vli id; 211 212 /// Pointer to function used to initialize the filter. 213 /// This is NULL to indicate end of array. 214 lzma_init_function init; 215 216 /// Pointer to filter's options structure 217 void *options; 218}; 219 220 221/// Hold data and function pointers of the next filter in the chain. 222struct lzma_next_coder_s { 223 /// Pointer to coder-specific data 224 void *coder; 225 226 /// Filter ID. This is LZMA_VLI_UNKNOWN when this structure doesn't 227 /// point to a filter coder. 228 lzma_vli id; 229 230 /// "Pointer" to init function. This is never called here. 231 /// We need only to detect if we are initializing a coder 232 /// that was allocated earlier. See lzma_next_coder_init and 233 /// lzma_next_strm_init macros in this file. 234 uintptr_t init; 235 236 /// Pointer to function to do the actual coding 237 lzma_code_function code; 238 239 /// Pointer to function to free lzma_next_coder.coder. This can 240 /// be NULL; in that case, lzma_free is called to free 241 /// lzma_next_coder.coder. 242 lzma_end_function end; 243 244 /// Pointer to a function to get progress information. If this is NULL, 245 /// lzma_stream.total_in and .total_out are used instead. 246 void (*get_progress)(void *coder, 247 uint64_t *progress_in, uint64_t *progress_out); 248 249 /// Pointer to function to return the type of the integrity check. 250 /// Most coders won't support this. 251 lzma_check (*get_check)(const void *coder); 252 253 /// Pointer to function to get and/or change the memory usage limit. 254 /// If new_memlimit == 0, the limit is not changed. 255 lzma_ret (*memconfig)(void *coder, uint64_t *memusage, 256 uint64_t *old_memlimit, uint64_t new_memlimit); 257 258 /// Update the filter-specific options or the whole filter chain 259 /// in the encoder. 260 lzma_ret (*update)(void *coder, const lzma_allocator *allocator, 261 const lzma_filter *filters, 262 const lzma_filter *reversed_filters); 263 264 /// Set how many bytes of output this coder may produce at maximum. 265 /// On success LZMA_OK must be returned. 266 /// If the filter chain as a whole cannot support this feature, 267 /// this must return LZMA_OPTIONS_ERROR. 268 /// If no input has been given to the coder and the requested limit 269 /// is too small, this must return LZMA_BUF_ERROR. If input has been 270 /// seen, LZMA_OK is allowed too. 271 lzma_ret (*set_out_limit)(void *coder, uint64_t *uncomp_size, 272 uint64_t out_limit); 273}; 274 275 276/// Macro to initialize lzma_next_coder structure 277#define LZMA_NEXT_CODER_INIT \ 278 (lzma_next_coder){ \ 279 .coder = NULL, \ 280 .init = (uintptr_t)(NULL), \ 281 .id = LZMA_VLI_UNKNOWN, \ 282 .code = NULL, \ 283 .end = NULL, \ 284 .get_progress = NULL, \ 285 .get_check = NULL, \ 286 .memconfig = NULL, \ 287 .update = NULL, \ 288 .set_out_limit = NULL, \ 289 } 290 291 292/// Internal data for lzma_strm_init, lzma_code, and lzma_end. A pointer to 293/// this is stored in lzma_stream. 294struct lzma_internal_s { 295 /// The actual coder that should do something useful 296 lzma_next_coder next; 297 298 /// Track the state of the coder. This is used to validate arguments 299 /// so that the actual coders can rely on e.g. that LZMA_SYNC_FLUSH 300 /// is used on every call to lzma_code until next.code has returned 301 /// LZMA_STREAM_END. 302 enum { 303 ISEQ_RUN, 304 ISEQ_SYNC_FLUSH, 305 ISEQ_FULL_FLUSH, 306 ISEQ_FINISH, 307 ISEQ_FULL_BARRIER, 308 ISEQ_END, 309 ISEQ_ERROR, 310 } sequence; 311 312 /// A copy of lzma_stream avail_in. This is used to verify that the 313 /// amount of input doesn't change once e.g. LZMA_FINISH has been 314 /// used. 315 size_t avail_in; 316 317 /// Indicates which lzma_action values are allowed by next.code. 318 bool supported_actions[LZMA_ACTION_MAX + 1]; 319 320 /// If true, lzma_code will return LZMA_BUF_ERROR if no progress was 321 /// made (no input consumed and no output produced by next.code). 322 bool allow_buf_error; 323}; 324 325 326/// Allocates memory 327lzma_attr_alloc_size(1) 328extern void *lzma_alloc(size_t size, const lzma_allocator *allocator); 329 330/// Allocates memory and zeroes it (like calloc()). This can be faster 331/// than lzma_alloc() + memzero() while being backward compatible with 332/// custom allocators. 333lzma_attr_alloc_size(1) 334extern void *lzma_alloc_zero(size_t size, const lzma_allocator *allocator); 335 336/// Frees memory 337extern void lzma_free(void *ptr, const lzma_allocator *allocator); 338 339 340/// Allocates strm->internal if it is NULL, and initializes *strm and 341/// strm->internal. This function is only called via lzma_next_strm_init macro. 342extern lzma_ret lzma_strm_init(lzma_stream *strm); 343 344/// Initializes the next filter in the chain, if any. This takes care of 345/// freeing the memory of previously initialized filter if it is different 346/// than the filter being initialized now. This way the actual filter 347/// initialization functions don't need to use lzma_next_coder_init macro. 348extern lzma_ret lzma_next_filter_init(lzma_next_coder *next, 349 const lzma_allocator *allocator, 350 const lzma_filter_info *filters); 351 352/// Update the next filter in the chain, if any. This checks that 353/// the application is not trying to change the Filter IDs. 354extern lzma_ret lzma_next_filter_update( 355 lzma_next_coder *next, const lzma_allocator *allocator, 356 const lzma_filter *reversed_filters); 357 358/// Frees the memory allocated for next->coder either using next->end or, 359/// if next->end is NULL, using lzma_free. 360extern void lzma_next_end(lzma_next_coder *next, 361 const lzma_allocator *allocator); 362 363 364/// Copy as much data as possible from in[] to out[] and update *in_pos 365/// and *out_pos accordingly. Returns the number of bytes copied. 366extern size_t lzma_bufcpy(const uint8_t *restrict in, size_t *restrict in_pos, 367 size_t in_size, uint8_t *restrict out, 368 size_t *restrict out_pos, size_t out_size); 369 370 371/// \brief Return if expression doesn't evaluate to LZMA_OK 372/// 373/// There are several situations where we want to return immediately 374/// with the value of expr if it isn't LZMA_OK. This macro shortens 375/// the code a little. 376#define return_if_error(expr) \ 377do { \ 378 const lzma_ret ret_ = (expr); \ 379 if (ret_ != LZMA_OK) \ 380 return ret_; \ 381} while (0) 382 383 384/// If next isn't already initialized, free the previous coder. Then mark 385/// that next is _possibly_ initialized for the coder using this macro. 386/// "Possibly" means that if e.g. allocation of next->coder fails, the 387/// structure isn't actually initialized for this coder, but leaving 388/// next->init to func is still OK. 389#define lzma_next_coder_init(func, next, allocator) \ 390do { \ 391 if ((uintptr_t)(func) != (next)->init) \ 392 lzma_next_end(next, allocator); \ 393 (next)->init = (uintptr_t)(func); \ 394} while (0) 395 396 397/// Initializes lzma_strm and calls func() to initialize strm->internal->next. 398/// (The function being called will use lzma_next_coder_init()). If 399/// initialization fails, memory that wasn't freed by func() is freed 400/// along strm->internal. 401#define lzma_next_strm_init(func, strm, ...) \ 402do { \ 403 return_if_error(lzma_strm_init(strm)); \ 404 const lzma_ret ret_ = func(&(strm)->internal->next, \ 405 (strm)->allocator, __VA_ARGS__); \ 406 if (ret_ != LZMA_OK) { \ 407 lzma_end(strm); \ 408 return ret_; \ 409 } \ 410} while (0) 411 412#endif 413