coder.c revision 207753
1/////////////////////////////////////////////////////////////////////////////// 2// 3/// \file coder.c 4/// \brief Compresses or uncompresses a file 5// 6// Author: Lasse Collin 7// 8// This file has been put into the public domain. 9// You can do whatever you want with this file. 10// 11/////////////////////////////////////////////////////////////////////////////// 12 13#include "private.h" 14 15 16/// Return value type for coder_init(). 17enum coder_init_ret { 18 CODER_INIT_NORMAL, 19 CODER_INIT_PASSTHRU, 20 CODER_INIT_ERROR, 21}; 22 23 24enum operation_mode opt_mode = MODE_COMPRESS; 25 26enum format_type opt_format = FORMAT_AUTO; 27 28 29/// Stream used to communicate with liblzma 30static lzma_stream strm = LZMA_STREAM_INIT; 31 32/// Filters needed for all encoding all formats, and also decoding in raw data 33static lzma_filter filters[LZMA_FILTERS_MAX + 1]; 34 35/// Input and output buffers 36static io_buf in_buf; 37static io_buf out_buf; 38 39/// Number of filters. Zero indicates that we are using a preset. 40static size_t filters_count = 0; 41 42/// Number of the preset (0-9) 43static size_t preset_number = 6; 44 45/// True if we should auto-adjust the compression settings to use less memory 46/// if memory usage limit is too low for the original settings. 47static bool auto_adjust = true; 48 49/// Indicate if no preset has been explicitly given. In that case, if we need 50/// to auto-adjust for lower memory usage, we won't print a warning. 51static bool preset_default = true; 52 53/// If a preset is used (no custom filter chain) and preset_extreme is true, 54/// a significantly slower compression is used to achieve slightly better 55/// compression ratio. 56static bool preset_extreme = false; 57 58/// Integrity check type 59static lzma_check check; 60 61/// This becomes false if the --check=CHECK option is used. 62static bool check_default = true; 63 64 65extern void 66coder_set_check(lzma_check new_check) 67{ 68 check = new_check; 69 check_default = false; 70 return; 71} 72 73 74extern void 75coder_set_preset(size_t new_preset) 76{ 77 preset_number = new_preset; 78 preset_default = false; 79 return; 80} 81 82 83extern void 84coder_set_extreme(void) 85{ 86 preset_extreme = true; 87 return; 88} 89 90 91extern void 92coder_add_filter(lzma_vli id, void *options) 93{ 94 if (filters_count == LZMA_FILTERS_MAX) 95 message_fatal(_("Maximum number of filters is four")); 96 97 filters[filters_count].id = id; 98 filters[filters_count].options = options; 99 ++filters_count; 100 101 return; 102} 103 104 105static void lzma_attribute((noreturn)) 106memlimit_too_small(uint64_t memory_usage) 107{ 108 message(V_ERROR, _("Memory usage limit is too low for the given " 109 "filter setup.")); 110 message_mem_needed(V_ERROR, memory_usage); 111 tuklib_exit(E_ERROR, E_ERROR, false); 112} 113 114 115extern void 116coder_set_compression_settings(void) 117{ 118 // Options for LZMA1 or LZMA2 in case we are using a preset. 119 static lzma_options_lzma opt_lzma; 120 121 if (filters_count == 0) { 122 // We are using a preset. This is not a good idea in raw mode 123 // except when playing around with things. Different versions 124 // of this software may use different options in presets, and 125 // thus make uncompressing the raw data difficult. 126 if (opt_format == FORMAT_RAW) { 127 // The message is shown only if warnings are allowed 128 // but the exit status isn't changed. 129 message(V_WARNING, _("Using a preset in raw mode " 130 "is discouraged.")); 131 message(V_WARNING, _("The exact options of the " 132 "presets may vary between software " 133 "versions.")); 134 } 135 136 // Get the preset for LZMA1 or LZMA2. 137 if (preset_extreme) 138 preset_number |= LZMA_PRESET_EXTREME; 139 140 if (lzma_lzma_preset(&opt_lzma, preset_number)) 141 message_bug(); 142 143 // Use LZMA2 except with --format=lzma we use LZMA1. 144 filters[0].id = opt_format == FORMAT_LZMA 145 ? LZMA_FILTER_LZMA1 : LZMA_FILTER_LZMA2; 146 filters[0].options = &opt_lzma; 147 filters_count = 1; 148 } else { 149 preset_default = false; 150 } 151 152 // Terminate the filter options array. 153 filters[filters_count].id = LZMA_VLI_UNKNOWN; 154 155 // If we are using the .lzma format, allow exactly one filter 156 // which has to be LZMA1. 157 if (opt_format == FORMAT_LZMA && (filters_count != 1 158 || filters[0].id != LZMA_FILTER_LZMA1)) 159 message_fatal(_("The .lzma format supports only " 160 "the LZMA1 filter")); 161 162 // If we are using the .xz format, make sure that there is no LZMA1 163 // filter to prevent LZMA_PROG_ERROR. 164 if (opt_format == FORMAT_XZ) 165 for (size_t i = 0; i < filters_count; ++i) 166 if (filters[i].id == LZMA_FILTER_LZMA1) 167 message_fatal(_("LZMA1 cannot be used " 168 "with the .xz format")); 169 170 // Print the selected filter chain. 171 message_filters(V_DEBUG, filters); 172 173 // If using --format=raw, we can be decoding. The memusage function 174 // also validates the filter chain and the options used for the 175 // filters. 176 const uint64_t memory_limit = hardware_memlimit_get(); 177 uint64_t memory_usage; 178 if (opt_mode == MODE_COMPRESS) 179 memory_usage = lzma_raw_encoder_memusage(filters); 180 else 181 memory_usage = lzma_raw_decoder_memusage(filters); 182 183 if (memory_usage == UINT64_MAX) 184 message_fatal(_("Unsupported filter chain or filter options")); 185 186 // Print memory usage info before possible dictionary 187 // size auto-adjusting. 188 message_mem_needed(V_DEBUG, memory_usage); 189 190 if (memory_usage > memory_limit) { 191 // If --no-auto-adjust was used or we didn't find LZMA1 or 192 // LZMA2 as the last filter, give an error immediately. 193 // --format=raw implies --no-auto-adjust. 194 if (!auto_adjust || opt_format == FORMAT_RAW) 195 memlimit_too_small(memory_usage); 196 197 assert(opt_mode == MODE_COMPRESS); 198 199 // Look for the last filter if it is LZMA2 or LZMA1, so 200 // we can make it use less RAM. With other filters we don't 201 // know what to do. 202 size_t i = 0; 203 while (filters[i].id != LZMA_FILTER_LZMA2 204 && filters[i].id != LZMA_FILTER_LZMA1) { 205 if (filters[i].id == LZMA_VLI_UNKNOWN) 206 memlimit_too_small(memory_usage); 207 208 ++i; 209 } 210 211 // Decrease the dictionary size until we meet the memory 212 // usage limit. First round down to full mebibytes. 213 lzma_options_lzma *opt = filters[i].options; 214 const uint32_t orig_dict_size = opt->dict_size; 215 opt->dict_size &= ~((UINT32_C(1) << 20) - 1); 216 while (true) { 217 // If it is below 1 MiB, auto-adjusting failed. We 218 // could be more sophisticated and scale it down even 219 // more, but let's see if many complain about this 220 // version. 221 // 222 // FIXME: Displays the scaled memory usage instead 223 // of the original. 224 if (opt->dict_size < (UINT32_C(1) << 20)) 225 memlimit_too_small(memory_usage); 226 227 memory_usage = lzma_raw_encoder_memusage(filters); 228 if (memory_usage == UINT64_MAX) 229 message_bug(); 230 231 // Accept it if it is low enough. 232 if (memory_usage <= memory_limit) 233 break; 234 235 // Otherwise 1 MiB down and try again. I hope this 236 // isn't too slow method for cases where the original 237 // dict_size is very big. 238 opt->dict_size -= UINT32_C(1) << 20; 239 } 240 241 // Tell the user that we decreased the dictionary size. 242 // However, omit the message if no preset or custom chain 243 // was given. FIXME: Always warn? 244 if (!preset_default) 245 message(V_WARNING, _("Adjusted LZMA%c dictionary size " 246 "from %s MiB to %s MiB to not exceed " 247 "the memory usage limit of %s MiB"), 248 filters[i].id == LZMA_FILTER_LZMA2 249 ? '2' : '1', 250 uint64_to_str(orig_dict_size >> 20, 0), 251 uint64_to_str(opt->dict_size >> 20, 1), 252 uint64_to_str(round_up_to_mib( 253 memory_limit), 2)); 254 } 255 256/* 257 // Limit the number of worker threads so that memory usage 258 // limit isn't exceeded. 259 assert(memory_usage > 0); 260 size_t thread_limit = memory_limit / memory_usage; 261 if (thread_limit == 0) 262 thread_limit = 1; 263 264 if (opt_threads > thread_limit) 265 opt_threads = thread_limit; 266*/ 267 268 if (check_default) { 269 // The default check type is CRC64, but fallback to CRC32 270 // if CRC64 isn't supported by the copy of liblzma we are 271 // using. CRC32 is always supported. 272 check = LZMA_CHECK_CRC64; 273 if (!lzma_check_is_supported(check)) 274 check = LZMA_CHECK_CRC32; 275 } 276 277 return; 278} 279 280 281/// Return true if the data in in_buf seems to be in the .xz format. 282static bool 283is_format_xz(void) 284{ 285 return strm.avail_in >= 6 && memcmp(in_buf.u8, "\3757zXZ", 6) == 0; 286} 287 288 289/// Return true if the data in in_buf seems to be in the .lzma format. 290static bool 291is_format_lzma(void) 292{ 293 // The .lzma header is 13 bytes. 294 if (strm.avail_in < 13) 295 return false; 296 297 // Decode the LZMA1 properties. 298 lzma_filter filter = { .id = LZMA_FILTER_LZMA1 }; 299 if (lzma_properties_decode(&filter, NULL, in_buf.u8, 5) != LZMA_OK) 300 return false; 301 302 // A hack to ditch tons of false positives: We allow only dictionary 303 // sizes that are 2^n or 2^n + 2^(n-1) or UINT32_MAX. LZMA_Alone 304 // created only files with 2^n, but accepts any dictionary size. 305 // If someone complains, this will be reconsidered. 306 lzma_options_lzma *opt = filter.options; 307 const uint32_t dict_size = opt->dict_size; 308 free(opt); 309 310 if (dict_size != UINT32_MAX) { 311 uint32_t d = dict_size - 1; 312 d |= d >> 2; 313 d |= d >> 3; 314 d |= d >> 4; 315 d |= d >> 8; 316 d |= d >> 16; 317 ++d; 318 if (d != dict_size || dict_size == 0) 319 return false; 320 } 321 322 // Another hack to ditch false positives: Assume that if the 323 // uncompressed size is known, it must be less than 256 GiB. 324 // Again, if someone complains, this will be reconsidered. 325 uint64_t uncompressed_size = 0; 326 for (size_t i = 0; i < 8; ++i) 327 uncompressed_size |= (uint64_t)(in_buf.u8[5 + i]) << (i * 8); 328 329 if (uncompressed_size != UINT64_MAX 330 && uncompressed_size > (UINT64_C(1) << 38)) 331 return false; 332 333 return true; 334} 335 336 337/// Detect the input file type (for now, this done only when decompressing), 338/// and initialize an appropriate coder. Return value indicates if a normal 339/// liblzma-based coder was initialized (CODER_INIT_NORMAL), if passthru 340/// mode should be used (CODER_INIT_PASSTHRU), or if an error occurred 341/// (CODER_INIT_ERROR). 342static enum coder_init_ret 343coder_init(file_pair *pair) 344{ 345 lzma_ret ret = LZMA_PROG_ERROR; 346 347 if (opt_mode == MODE_COMPRESS) { 348 switch (opt_format) { 349 case FORMAT_AUTO: 350 // args.c ensures this. 351 assert(0); 352 break; 353 354 case FORMAT_XZ: 355 ret = lzma_stream_encoder(&strm, filters, check); 356 break; 357 358 case FORMAT_LZMA: 359 ret = lzma_alone_encoder(&strm, filters[0].options); 360 break; 361 362 case FORMAT_RAW: 363 ret = lzma_raw_encoder(&strm, filters); 364 break; 365 } 366 } else { 367 const uint32_t flags = LZMA_TELL_UNSUPPORTED_CHECK 368 | LZMA_CONCATENATED; 369 370 // We abuse FORMAT_AUTO to indicate unknown file format, 371 // for which we may consider passthru mode. 372 enum format_type init_format = FORMAT_AUTO; 373 374 switch (opt_format) { 375 case FORMAT_AUTO: 376 if (is_format_xz()) 377 init_format = FORMAT_XZ; 378 else if (is_format_lzma()) 379 init_format = FORMAT_LZMA; 380 break; 381 382 case FORMAT_XZ: 383 if (is_format_xz()) 384 init_format = FORMAT_XZ; 385 break; 386 387 case FORMAT_LZMA: 388 if (is_format_lzma()) 389 init_format = FORMAT_LZMA; 390 break; 391 392 case FORMAT_RAW: 393 init_format = FORMAT_RAW; 394 break; 395 } 396 397 switch (init_format) { 398 case FORMAT_AUTO: 399 // Uknown file format. If --decompress --stdout 400 // --force have been given, then we copy the input 401 // as is to stdout. Checking for MODE_DECOMPRESS 402 // is needed, because we don't want to do use 403 // passthru mode with --test. 404 if (opt_mode == MODE_DECOMPRESS 405 && opt_stdout && opt_force) 406 return CODER_INIT_PASSTHRU; 407 408 ret = LZMA_FORMAT_ERROR; 409 break; 410 411 case FORMAT_XZ: 412 ret = lzma_stream_decoder(&strm, 413 hardware_memlimit_get(), flags); 414 break; 415 416 case FORMAT_LZMA: 417 ret = lzma_alone_decoder(&strm, 418 hardware_memlimit_get()); 419 break; 420 421 case FORMAT_RAW: 422 // Memory usage has already been checked in 423 // coder_set_compression_settings(). 424 ret = lzma_raw_decoder(&strm, filters); 425 break; 426 } 427 428 // Try to decode the headers. This will catch too low 429 // memory usage limit in case it happens in the first 430 // Block of the first Stream, which is where it very 431 // probably will happen if it is going to happen. 432 if (ret == LZMA_OK && init_format != FORMAT_RAW) { 433 strm.next_out = NULL; 434 strm.avail_out = 0; 435 ret = lzma_code(&strm, LZMA_RUN); 436 } 437 } 438 439 if (ret != LZMA_OK) { 440 message_error("%s: %s", pair->src_name, message_strm(ret)); 441 if (ret == LZMA_MEMLIMIT_ERROR) 442 message_mem_needed(V_ERROR, lzma_memusage(&strm)); 443 444 return CODER_INIT_ERROR; 445 } 446 447 return CODER_INIT_NORMAL; 448} 449 450 451/// Compress or decompress using liblzma. 452static bool 453coder_normal(file_pair *pair) 454{ 455 // Encoder needs to know when we have given all the input to it. 456 // The decoders need to know it too when we are using 457 // LZMA_CONCATENATED. We need to check for src_eof here, because 458 // the first input chunk has been already read, and that may 459 // have been the only chunk we will read. 460 lzma_action action = pair->src_eof ? LZMA_FINISH : LZMA_RUN; 461 462 lzma_ret ret; 463 464 // Assume that something goes wrong. 465 bool success = false; 466 467 strm.next_out = out_buf.u8; 468 strm.avail_out = IO_BUFFER_SIZE; 469 470 while (!user_abort) { 471 // Fill the input buffer if it is empty and we haven't reached 472 // end of file yet. 473 if (strm.avail_in == 0 && !pair->src_eof) { 474 strm.next_in = in_buf.u8; 475 strm.avail_in = io_read( 476 pair, &in_buf, IO_BUFFER_SIZE); 477 478 if (strm.avail_in == SIZE_MAX) 479 break; 480 481 if (pair->src_eof) 482 action = LZMA_FINISH; 483 } 484 485 // Let liblzma do the actual work. 486 ret = lzma_code(&strm, action); 487 488 // Write out if the output buffer became full. 489 if (strm.avail_out == 0) { 490 if (opt_mode != MODE_TEST && io_write(pair, &out_buf, 491 IO_BUFFER_SIZE - strm.avail_out)) 492 break; 493 494 strm.next_out = out_buf.u8; 495 strm.avail_out = IO_BUFFER_SIZE; 496 } 497 498 if (ret != LZMA_OK) { 499 // Determine if the return value indicates that we 500 // won't continue coding. 501 const bool stop = ret != LZMA_NO_CHECK 502 && ret != LZMA_UNSUPPORTED_CHECK; 503 504 if (stop) { 505 // Write the remaining bytes even if something 506 // went wrong, because that way the user gets 507 // as much data as possible, which can be good 508 // when trying to get at least some useful 509 // data out of damaged files. 510 if (opt_mode != MODE_TEST && io_write(pair, 511 &out_buf, IO_BUFFER_SIZE 512 - strm.avail_out)) 513 break; 514 } 515 516 if (ret == LZMA_STREAM_END) { 517 // Check that there is no trailing garbage. 518 // This is needed for LZMA_Alone and raw 519 // streams. 520 if (strm.avail_in == 0 && !pair->src_eof) { 521 // Try reading one more byte. 522 // Hopefully we don't get any more 523 // input, and thus pair->src_eof 524 // becomes true. 525 strm.avail_in = io_read( 526 pair, &in_buf, 1); 527 if (strm.avail_in == SIZE_MAX) 528 break; 529 530 assert(strm.avail_in == 0 531 || strm.avail_in == 1); 532 } 533 534 if (strm.avail_in == 0) { 535 assert(pair->src_eof); 536 success = true; 537 break; 538 } 539 540 // We hadn't reached the end of the file. 541 ret = LZMA_DATA_ERROR; 542 assert(stop); 543 } 544 545 // If we get here and stop is true, something went 546 // wrong and we print an error. Otherwise it's just 547 // a warning and coding can continue. 548 if (stop) { 549 message_error("%s: %s", pair->src_name, 550 message_strm(ret)); 551 } else { 552 message_warning("%s: %s", pair->src_name, 553 message_strm(ret)); 554 555 // When compressing, all possible errors set 556 // stop to true. 557 assert(opt_mode != MODE_COMPRESS); 558 } 559 560 if (ret == LZMA_MEMLIMIT_ERROR) { 561 // Display how much memory it would have 562 // actually needed. 563 message_mem_needed(V_ERROR, 564 lzma_memusage(&strm)); 565 } 566 567 if (stop) 568 break; 569 } 570 571 // Show progress information under certain conditions. 572 message_progress_update(); 573 } 574 575 return success; 576} 577 578 579/// Copy from input file to output file without processing the data in any 580/// way. This is used only when trying to decompress unrecognized files 581/// with --decompress --stdout --force, so the output is always stdout. 582static bool 583coder_passthru(file_pair *pair) 584{ 585 while (strm.avail_in != 0) { 586 if (user_abort) 587 return false; 588 589 if (io_write(pair, &in_buf, strm.avail_in)) 590 return false; 591 592 strm.total_in += strm.avail_in; 593 strm.total_out = strm.total_in; 594 message_progress_update(); 595 596 strm.avail_in = io_read(pair, &in_buf, IO_BUFFER_SIZE); 597 if (strm.avail_in == SIZE_MAX) 598 return false; 599 } 600 601 return true; 602} 603 604 605extern void 606coder_run(const char *filename) 607{ 608 // Set and possibly print the filename for the progress message. 609 message_filename(filename); 610 611 // Try to open the input file. 612 file_pair *pair = io_open_src(filename); 613 if (pair == NULL) 614 return; 615 616 // Assume that something goes wrong. 617 bool success = false; 618 619 // Read the first chunk of input data. This is needed to detect 620 // the input file type (for now, only for decompression). 621 strm.next_in = in_buf.u8; 622 strm.avail_in = io_read(pair, &in_buf, IO_BUFFER_SIZE); 623 624 if (strm.avail_in != SIZE_MAX) { 625 // Initialize the coder. This will detect the file format 626 // and, in decompression or testing mode, check the memory 627 // usage of the first Block too. This way we don't try to 628 // open the destination file if we see that coding wouldn't 629 // work at all anyway. This also avoids deleting the old 630 // "target" file if --force was used. 631 const enum coder_init_ret init_ret = coder_init(pair); 632 633 if (init_ret != CODER_INIT_ERROR && !user_abort) { 634 // Don't open the destination file when --test 635 // is used. 636 if (opt_mode == MODE_TEST || !io_open_dest(pair)) { 637 // Initialize the progress indicator. 638 const uint64_t in_size 639 = pair->src_st.st_size <= 0 640 ? 0 : pair->src_st.st_size; 641 message_progress_start(&strm, in_size); 642 643 // Do the actual coding or passthru. 644 if (init_ret == CODER_INIT_NORMAL) 645 success = coder_normal(pair); 646 else 647 success = coder_passthru(pair); 648 649 message_progress_end(success); 650 } 651 } 652 } 653 654 // Close the file pair. It needs to know if coding was successful to 655 // know if the source or target file should be unlinked. 656 io_close(pair, success); 657 658 return; 659} 660