coder.c revision 274261
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; 25enum format_type opt_format = FORMAT_AUTO; 26bool opt_auto_adjust = true; 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 uint32_t filters_count = 0; 41 42/// Number of the preset (0-9) 43static uint32_t preset_number = LZMA_PRESET_DEFAULT; 44 45/// Integrity check type 46static lzma_check check; 47 48/// This becomes false if the --check=CHECK option is used. 49static bool check_default = true; 50 51 52extern void 53coder_set_check(lzma_check new_check) 54{ 55 check = new_check; 56 check_default = false; 57 return; 58} 59 60 61static void 62forget_filter_chain(void) 63{ 64 // Setting a preset makes us forget a possibly defined custom 65 // filter chain. 66 while (filters_count > 0) { 67 --filters_count; 68 free(filters[filters_count].options); 69 filters[filters_count].options = NULL; 70 } 71 72 return; 73} 74 75 76extern void 77coder_set_preset(uint32_t new_preset) 78{ 79 preset_number &= ~LZMA_PRESET_LEVEL_MASK; 80 preset_number |= new_preset; 81 forget_filter_chain(); 82 return; 83} 84 85 86extern void 87coder_set_extreme(void) 88{ 89 preset_number |= LZMA_PRESET_EXTREME; 90 forget_filter_chain(); 91 return; 92} 93 94 95extern void 96coder_add_filter(lzma_vli id, void *options) 97{ 98 if (filters_count == LZMA_FILTERS_MAX) 99 message_fatal(_("Maximum number of filters is four")); 100 101 filters[filters_count].id = id; 102 filters[filters_count].options = options; 103 ++filters_count; 104 105 // Setting a custom filter chain makes us forget the preset options. 106 // This makes a difference if one specifies e.g. "xz -9 --lzma2 -e" 107 // where the custom filter chain resets the preset level back to 108 // the default 6, making the example equivalent to "xz -6e". 109 preset_number = LZMA_PRESET_DEFAULT; 110 111 return; 112} 113 114 115static void lzma_attribute((__noreturn__)) 116memlimit_too_small(uint64_t memory_usage) 117{ 118 message(V_ERROR, _("Memory usage limit is too low for the given " 119 "filter setup.")); 120 message_mem_needed(V_ERROR, memory_usage); 121 tuklib_exit(E_ERROR, E_ERROR, false); 122} 123 124 125extern void 126coder_set_compression_settings(void) 127{ 128 // Options for LZMA1 or LZMA2 in case we are using a preset. 129 static lzma_options_lzma opt_lzma; 130 131 if (filters_count == 0) { 132 // We are using a preset. This is not a good idea in raw mode 133 // except when playing around with things. Different versions 134 // of this software may use different options in presets, and 135 // thus make uncompressing the raw data difficult. 136 if (opt_format == FORMAT_RAW) { 137 // The message is shown only if warnings are allowed 138 // but the exit status isn't changed. 139 message(V_WARNING, _("Using a preset in raw mode " 140 "is discouraged.")); 141 message(V_WARNING, _("The exact options of the " 142 "presets may vary between software " 143 "versions.")); 144 } 145 146 // Get the preset for LZMA1 or LZMA2. 147 if (lzma_lzma_preset(&opt_lzma, preset_number)) 148 message_bug(); 149 150 // Use LZMA2 except with --format=lzma we use LZMA1. 151 filters[0].id = opt_format == FORMAT_LZMA 152 ? LZMA_FILTER_LZMA1 : LZMA_FILTER_LZMA2; 153 filters[0].options = &opt_lzma; 154 filters_count = 1; 155 } 156 157 // Terminate the filter options array. 158 filters[filters_count].id = LZMA_VLI_UNKNOWN; 159 160 // If we are using the .lzma format, allow exactly one filter 161 // which has to be LZMA1. 162 if (opt_format == FORMAT_LZMA && (filters_count != 1 163 || filters[0].id != LZMA_FILTER_LZMA1)) 164 message_fatal(_("The .lzma format supports only " 165 "the LZMA1 filter")); 166 167 // If we are using the .xz format, make sure that there is no LZMA1 168 // filter to prevent LZMA_PROG_ERROR. 169 if (opt_format == FORMAT_XZ) 170 for (size_t i = 0; i < filters_count; ++i) 171 if (filters[i].id == LZMA_FILTER_LZMA1) 172 message_fatal(_("LZMA1 cannot be used " 173 "with the .xz format")); 174 175 // Print the selected filter chain. 176 message_filters_show(V_DEBUG, filters); 177 178 // If using --format=raw, we can be decoding. The memusage function 179 // also validates the filter chain and the options used for the 180 // filters. 181 const uint64_t memory_limit = hardware_memlimit_get(opt_mode); 182 uint64_t memory_usage; 183 if (opt_mode == MODE_COMPRESS) 184 memory_usage = lzma_raw_encoder_memusage(filters); 185 else 186 memory_usage = lzma_raw_decoder_memusage(filters); 187 188 if (memory_usage == UINT64_MAX) 189 message_fatal(_("Unsupported filter chain or filter options")); 190 191 // Print memory usage info before possible dictionary 192 // size auto-adjusting. 193 message_mem_needed(V_DEBUG, memory_usage); 194 if (opt_mode == MODE_COMPRESS) { 195 const uint64_t decmem = lzma_raw_decoder_memusage(filters); 196 if (decmem != UINT64_MAX) 197 message(V_DEBUG, _("Decompression will need " 198 "%s MiB of memory."), uint64_to_str( 199 round_up_to_mib(decmem), 0)); 200 } 201 202 if (memory_usage > memory_limit) { 203 // If --no-adjust was used or we didn't find LZMA1 or 204 // LZMA2 as the last filter, give an error immediately. 205 // --format=raw implies --no-adjust. 206 if (!opt_auto_adjust || opt_format == FORMAT_RAW) 207 memlimit_too_small(memory_usage); 208 209 assert(opt_mode == MODE_COMPRESS); 210 211 // Look for the last filter if it is LZMA2 or LZMA1, so 212 // we can make it use less RAM. With other filters we don't 213 // know what to do. 214 size_t i = 0; 215 while (filters[i].id != LZMA_FILTER_LZMA2 216 && filters[i].id != LZMA_FILTER_LZMA1) { 217 if (filters[i].id == LZMA_VLI_UNKNOWN) 218 memlimit_too_small(memory_usage); 219 220 ++i; 221 } 222 223 // Decrease the dictionary size until we meet the memory 224 // usage limit. First round down to full mebibytes. 225 lzma_options_lzma *opt = filters[i].options; 226 const uint32_t orig_dict_size = opt->dict_size; 227 opt->dict_size &= ~((UINT32_C(1) << 20) - 1); 228 while (true) { 229 // If it is below 1 MiB, auto-adjusting failed. We 230 // could be more sophisticated and scale it down even 231 // more, but let's see if many complain about this 232 // version. 233 // 234 // FIXME: Displays the scaled memory usage instead 235 // of the original. 236 if (opt->dict_size < (UINT32_C(1) << 20)) 237 memlimit_too_small(memory_usage); 238 239 memory_usage = lzma_raw_encoder_memusage(filters); 240 if (memory_usage == UINT64_MAX) 241 message_bug(); 242 243 // Accept it if it is low enough. 244 if (memory_usage <= memory_limit) 245 break; 246 247 // Otherwise 1 MiB down and try again. I hope this 248 // isn't too slow method for cases where the original 249 // dict_size is very big. 250 opt->dict_size -= UINT32_C(1) << 20; 251 } 252 253 // Tell the user that we decreased the dictionary size. 254 message(V_WARNING, _("Adjusted LZMA%c dictionary size " 255 "from %s MiB to %s MiB to not exceed " 256 "the memory usage limit of %s MiB"), 257 filters[i].id == LZMA_FILTER_LZMA2 258 ? '2' : '1', 259 uint64_to_str(orig_dict_size >> 20, 0), 260 uint64_to_str(opt->dict_size >> 20, 1), 261 uint64_to_str(round_up_to_mib( 262 memory_limit), 2)); 263 } 264 265/* 266 // Limit the number of worker threads so that memory usage 267 // limit isn't exceeded. 268 assert(memory_usage > 0); 269 size_t thread_limit = memory_limit / memory_usage; 270 if (thread_limit == 0) 271 thread_limit = 1; 272 273 if (opt_threads > thread_limit) 274 opt_threads = thread_limit; 275*/ 276 277 if (check_default) { 278 // The default check type is CRC64, but fallback to CRC32 279 // if CRC64 isn't supported by the copy of liblzma we are 280 // using. CRC32 is always supported. 281 check = LZMA_CHECK_CRC64; 282 if (!lzma_check_is_supported(check)) 283 check = LZMA_CHECK_CRC32; 284 } 285 286 return; 287} 288 289 290/// Return true if the data in in_buf seems to be in the .xz format. 291static bool 292is_format_xz(void) 293{ 294 // Specify the magic as hex to be compatible with EBCDIC systems. 295 static const uint8_t magic[6] = { 0xFD, 0x37, 0x7A, 0x58, 0x5A, 0x00 }; 296 return strm.avail_in >= sizeof(magic) 297 && memcmp(in_buf.u8, magic, sizeof(magic)) == 0; 298} 299 300 301/// Return true if the data in in_buf seems to be in the .lzma format. 302static bool 303is_format_lzma(void) 304{ 305 // The .lzma header is 13 bytes. 306 if (strm.avail_in < 13) 307 return false; 308 309 // Decode the LZMA1 properties. 310 lzma_filter filter = { .id = LZMA_FILTER_LZMA1 }; 311 if (lzma_properties_decode(&filter, NULL, in_buf.u8, 5) != LZMA_OK) 312 return false; 313 314 // A hack to ditch tons of false positives: We allow only dictionary 315 // sizes that are 2^n or 2^n + 2^(n-1) or UINT32_MAX. LZMA_Alone 316 // created only files with 2^n, but accepts any dictionary size. 317 // If someone complains, this will be reconsidered. 318 lzma_options_lzma *opt = filter.options; 319 const uint32_t dict_size = opt->dict_size; 320 free(opt); 321 322 if (dict_size != UINT32_MAX) { 323 uint32_t d = dict_size - 1; 324 d |= d >> 2; 325 d |= d >> 3; 326 d |= d >> 4; 327 d |= d >> 8; 328 d |= d >> 16; 329 ++d; 330 if (d != dict_size || dict_size == 0) 331 return false; 332 } 333 334 // Another hack to ditch false positives: Assume that if the 335 // uncompressed size is known, it must be less than 256 GiB. 336 // Again, if someone complains, this will be reconsidered. 337 uint64_t uncompressed_size = 0; 338 for (size_t i = 0; i < 8; ++i) 339 uncompressed_size |= (uint64_t)(in_buf.u8[5 + i]) << (i * 8); 340 341 if (uncompressed_size != UINT64_MAX 342 && uncompressed_size > (UINT64_C(1) << 38)) 343 return false; 344 345 return true; 346} 347 348 349/// Detect the input file type (for now, this done only when decompressing), 350/// and initialize an appropriate coder. Return value indicates if a normal 351/// liblzma-based coder was initialized (CODER_INIT_NORMAL), if passthru 352/// mode should be used (CODER_INIT_PASSTHRU), or if an error occurred 353/// (CODER_INIT_ERROR). 354static enum coder_init_ret 355coder_init(file_pair *pair) 356{ 357 lzma_ret ret = LZMA_PROG_ERROR; 358 359 if (opt_mode == MODE_COMPRESS) { 360 switch (opt_format) { 361 case FORMAT_AUTO: 362 // args.c ensures this. 363 assert(0); 364 break; 365 366 case FORMAT_XZ: 367 ret = lzma_stream_encoder(&strm, filters, check); 368 break; 369 370 case FORMAT_LZMA: 371 ret = lzma_alone_encoder(&strm, filters[0].options); 372 break; 373 374 case FORMAT_RAW: 375 ret = lzma_raw_encoder(&strm, filters); 376 break; 377 } 378 } else { 379 const uint32_t flags = LZMA_TELL_UNSUPPORTED_CHECK 380 | LZMA_CONCATENATED; 381 382 // We abuse FORMAT_AUTO to indicate unknown file format, 383 // for which we may consider passthru mode. 384 enum format_type init_format = FORMAT_AUTO; 385 386 switch (opt_format) { 387 case FORMAT_AUTO: 388 if (is_format_xz()) 389 init_format = FORMAT_XZ; 390 else if (is_format_lzma()) 391 init_format = FORMAT_LZMA; 392 break; 393 394 case FORMAT_XZ: 395 if (is_format_xz()) 396 init_format = FORMAT_XZ; 397 break; 398 399 case FORMAT_LZMA: 400 if (is_format_lzma()) 401 init_format = FORMAT_LZMA; 402 break; 403 404 case FORMAT_RAW: 405 init_format = FORMAT_RAW; 406 break; 407 } 408 409 switch (init_format) { 410 case FORMAT_AUTO: 411 // Uknown file format. If --decompress --stdout 412 // --force have been given, then we copy the input 413 // as is to stdout. Checking for MODE_DECOMPRESS 414 // is needed, because we don't want to do use 415 // passthru mode with --test. 416 if (opt_mode == MODE_DECOMPRESS 417 && opt_stdout && opt_force) 418 return CODER_INIT_PASSTHRU; 419 420 ret = LZMA_FORMAT_ERROR; 421 break; 422 423 case FORMAT_XZ: 424 ret = lzma_stream_decoder(&strm, 425 hardware_memlimit_get( 426 MODE_DECOMPRESS), flags); 427 break; 428 429 case FORMAT_LZMA: 430 ret = lzma_alone_decoder(&strm, 431 hardware_memlimit_get( 432 MODE_DECOMPRESS)); 433 break; 434 435 case FORMAT_RAW: 436 // Memory usage has already been checked in 437 // coder_set_compression_settings(). 438 ret = lzma_raw_decoder(&strm, filters); 439 break; 440 } 441 442 // Try to decode the headers. This will catch too low 443 // memory usage limit in case it happens in the first 444 // Block of the first Stream, which is where it very 445 // probably will happen if it is going to happen. 446 if (ret == LZMA_OK && init_format != FORMAT_RAW) { 447 strm.next_out = NULL; 448 strm.avail_out = 0; 449 ret = lzma_code(&strm, LZMA_RUN); 450 } 451 } 452 453 if (ret != LZMA_OK) { 454 message_error("%s: %s", pair->src_name, message_strm(ret)); 455 if (ret == LZMA_MEMLIMIT_ERROR) 456 message_mem_needed(V_ERROR, lzma_memusage(&strm)); 457 458 return CODER_INIT_ERROR; 459 } 460 461 return CODER_INIT_NORMAL; 462} 463 464 465/// Compress or decompress using liblzma. 466static bool 467coder_normal(file_pair *pair) 468{ 469 // Encoder needs to know when we have given all the input to it. 470 // The decoders need to know it too when we are using 471 // LZMA_CONCATENATED. We need to check for src_eof here, because 472 // the first input chunk has been already read, and that may 473 // have been the only chunk we will read. 474 lzma_action action = pair->src_eof ? LZMA_FINISH : LZMA_RUN; 475 476 lzma_ret ret; 477 478 // Assume that something goes wrong. 479 bool success = false; 480 481 strm.next_out = out_buf.u8; 482 strm.avail_out = IO_BUFFER_SIZE; 483 484 while (!user_abort) { 485 // Fill the input buffer if it is empty and we haven't reached 486 // end of file yet. 487 if (strm.avail_in == 0 && !pair->src_eof) { 488 strm.next_in = in_buf.u8; 489 strm.avail_in = io_read( 490 pair, &in_buf, IO_BUFFER_SIZE); 491 492 if (strm.avail_in == SIZE_MAX) 493 break; 494 495 if (pair->src_eof) 496 action = LZMA_FINISH; 497 } 498 499 // Let liblzma do the actual work. 500 ret = lzma_code(&strm, action); 501 502 // Write out if the output buffer became full. 503 if (strm.avail_out == 0) { 504 if (opt_mode != MODE_TEST && io_write(pair, &out_buf, 505 IO_BUFFER_SIZE - strm.avail_out)) 506 break; 507 508 strm.next_out = out_buf.u8; 509 strm.avail_out = IO_BUFFER_SIZE; 510 } 511 512 if (ret != LZMA_OK) { 513 // Determine if the return value indicates that we 514 // won't continue coding. 515 const bool stop = ret != LZMA_NO_CHECK 516 && ret != LZMA_UNSUPPORTED_CHECK; 517 518 if (stop) { 519 // Write the remaining bytes even if something 520 // went wrong, because that way the user gets 521 // as much data as possible, which can be good 522 // when trying to get at least some useful 523 // data out of damaged files. 524 if (opt_mode != MODE_TEST && io_write(pair, 525 &out_buf, IO_BUFFER_SIZE 526 - strm.avail_out)) 527 break; 528 } 529 530 if (ret == LZMA_STREAM_END) { 531 // Check that there is no trailing garbage. 532 // This is needed for LZMA_Alone and raw 533 // streams. 534 if (strm.avail_in == 0 && !pair->src_eof) { 535 // Try reading one more byte. 536 // Hopefully we don't get any more 537 // input, and thus pair->src_eof 538 // becomes true. 539 strm.avail_in = io_read( 540 pair, &in_buf, 1); 541 if (strm.avail_in == SIZE_MAX) 542 break; 543 544 assert(strm.avail_in == 0 545 || strm.avail_in == 1); 546 } 547 548 if (strm.avail_in == 0) { 549 assert(pair->src_eof); 550 success = true; 551 break; 552 } 553 554 // We hadn't reached the end of the file. 555 ret = LZMA_DATA_ERROR; 556 assert(stop); 557 } 558 559 // If we get here and stop is true, something went 560 // wrong and we print an error. Otherwise it's just 561 // a warning and coding can continue. 562 if (stop) { 563 message_error("%s: %s", pair->src_name, 564 message_strm(ret)); 565 } else { 566 message_warning("%s: %s", pair->src_name, 567 message_strm(ret)); 568 569 // When compressing, all possible errors set 570 // stop to true. 571 assert(opt_mode != MODE_COMPRESS); 572 } 573 574 if (ret == LZMA_MEMLIMIT_ERROR) { 575 // Display how much memory it would have 576 // actually needed. 577 message_mem_needed(V_ERROR, 578 lzma_memusage(&strm)); 579 } 580 581 if (stop) 582 break; 583 } 584 585 // Show progress information under certain conditions. 586 message_progress_update(); 587 } 588 589 return success; 590} 591 592 593/// Copy from input file to output file without processing the data in any 594/// way. This is used only when trying to decompress unrecognized files 595/// with --decompress --stdout --force, so the output is always stdout. 596static bool 597coder_passthru(file_pair *pair) 598{ 599 while (strm.avail_in != 0) { 600 if (user_abort) 601 return false; 602 603 if (io_write(pair, &in_buf, strm.avail_in)) 604 return false; 605 606 strm.total_in += strm.avail_in; 607 strm.total_out = strm.total_in; 608 message_progress_update(); 609 610 strm.avail_in = io_read(pair, &in_buf, IO_BUFFER_SIZE); 611 if (strm.avail_in == SIZE_MAX) 612 return false; 613 } 614 615 return true; 616} 617 618 619extern void 620coder_run(const char *filename) 621{ 622 // Set and possibly print the filename for the progress message. 623 message_filename(filename); 624 625 // Try to open the input file. 626 file_pair *pair = io_open_src(filename); 627 if (pair == NULL) 628 return; 629 630 // Assume that something goes wrong. 631 bool success = false; 632 633 // Read the first chunk of input data. This is needed to detect 634 // the input file type (for now, only for decompression). 635 strm.next_in = in_buf.u8; 636 strm.avail_in = io_read(pair, &in_buf, IO_BUFFER_SIZE); 637 638 if (strm.avail_in != SIZE_MAX) { 639 // Initialize the coder. This will detect the file format 640 // and, in decompression or testing mode, check the memory 641 // usage of the first Block too. This way we don't try to 642 // open the destination file if we see that coding wouldn't 643 // work at all anyway. This also avoids deleting the old 644 // "target" file if --force was used. 645 const enum coder_init_ret init_ret = coder_init(pair); 646 647 if (init_ret != CODER_INIT_ERROR && !user_abort) { 648 // Don't open the destination file when --test 649 // is used. 650 if (opt_mode == MODE_TEST || !io_open_dest(pair)) { 651 // Initialize the progress indicator. 652 const uint64_t in_size 653 = pair->src_st.st_size <= 0 654 ? 0 : pair->src_st.st_size; 655 message_progress_start(&strm, in_size); 656 657 // Do the actual coding or passthru. 658 if (init_ret == CODER_INIT_NORMAL) 659 success = coder_normal(pair); 660 else 661 success = coder_passthru(pair); 662 663 message_progress_end(success); 664 } 665 } 666 } 667 668 // Close the file pair. It needs to know if coding was successful to 669 // know if the source or target file should be unlinked. 670 io_close(pair, success); 671 672 return; 673} 674