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(). 17 enum coder_init_ret { 18 CODER_INIT_NORMAL, 19 CODER_INIT_PASSTHRU, 20 CODER_INIT_ERROR, 21 }; 22 23 24 enum operation_mode opt_mode = MODE_COMPRESS; 25 enum format_type opt_format = FORMAT_AUTO; 26 bool opt_auto_adjust = true; 27 28 29 /// Stream used to communicate with liblzma 30 static lzma_stream strm = LZMA_STREAM_INIT; 31 32 /// Filters needed for all encoding all formats, and also decoding in raw data 33 static lzma_filter filters[LZMA_FILTERS_MAX + 1]; 34 35 /// Input and output buffers 36 static io_buf in_buf; 37 static io_buf out_buf; 38 39 /// Number of filters. Zero indicates that we are using a preset. 40 static size_t filters_count = 0; 41 42 /// Number of the preset (0-9) 43 static size_t preset_number = 6; 44 45 /// If a preset is used (no custom filter chain) and preset_extreme is true, 46 /// a significantly slower compression is used to achieve slightly better 47 /// compression ratio. 48 static bool preset_extreme = false; 49 50 /// Integrity check type 51 static lzma_check check; 52 53 /// This becomes false if the --check=CHECK option is used. 54 static bool check_default = true; 55 56 57 extern void 58 coder_set_check(lzma_check new_check) 59 { 60 check = new_check; 61 check_default = false; 62 return; 63 } 64 65 66 extern void 67 coder_set_preset(size_t new_preset) 68 { 69 preset_number = new_preset; 70 71 // Setting a preset makes us forget a possibly defined custom 72 // filter chain. 73 while (filters_count > 0) { 74 --filters_count; 75 free(filters[filters_count].options); 76 filters[filters_count].options = NULL; 77 } 78 79 return; 80 } 81 82 83 extern void 84 coder_set_extreme(void) 85 { 86 preset_extreme = true; 87 return; 88 } 89 90 91 extern void 92 coder_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 } 103 104 105 static void lzma_attribute((__noreturn__)) 106 memlimit_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 } 113 114 115 extern void 116 coder_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 } 149 150 // Terminate the filter options array. 151 filters[filters_count].id = LZMA_VLI_UNKNOWN; 152 153 // If we are using the .lzma format, allow exactly one filter 154 // which has to be LZMA1. 155 if (opt_format == FORMAT_LZMA && (filters_count != 1 156 || filters[0].id != LZMA_FILTER_LZMA1)) 157 message_fatal(_("The .lzma format supports only " 158 "the LZMA1 filter")); 159 160 // If we are using the .xz format, make sure that there is no LZMA1 161 // filter to prevent LZMA_PROG_ERROR. 162 if (opt_format == FORMAT_XZ) 163 for (size_t i = 0; i < filters_count; ++i) 164 if (filters[i].id == LZMA_FILTER_LZMA1) 165 message_fatal(_("LZMA1 cannot be used " 166 "with the .xz format")); 167 168 // Print the selected filter chain. 169 message_filters_show(V_DEBUG, filters); 170 171 // If using --format=raw, we can be decoding. The memusage function 172 // also validates the filter chain and the options used for the 173 // filters. 174 const uint64_t memory_limit = hardware_memlimit_get(opt_mode); 175 uint64_t memory_usage; 176 if (opt_mode == MODE_COMPRESS) 177 memory_usage = lzma_raw_encoder_memusage(filters); 178 else 179 memory_usage = lzma_raw_decoder_memusage(filters); 180 181 if (memory_usage == UINT64_MAX) 182 message_fatal(_("Unsupported filter chain or filter options")); 183 184 // Print memory usage info before possible dictionary 185 // size auto-adjusting. 186 message_mem_needed(V_DEBUG, memory_usage); 187 if (opt_mode == MODE_COMPRESS) { 188 const uint64_t decmem = lzma_raw_decoder_memusage(filters); 189 if (decmem != UINT64_MAX) 190 message(V_DEBUG, _("Decompression will need " 191 "%s MiB of memory."), uint64_to_str( 192 round_up_to_mib(decmem), 0)); 193 } 194 195 if (memory_usage > memory_limit) { 196 // If --no-auto-adjust was used or we didn't find LZMA1 or 197 // LZMA2 as the last filter, give an error immediately. 198 // --format=raw implies --no-auto-adjust. 199 if (!opt_auto_adjust || opt_format == FORMAT_RAW) 200 memlimit_too_small(memory_usage); 201 202 assert(opt_mode == MODE_COMPRESS); 203 204 // Look for the last filter if it is LZMA2 or LZMA1, so 205 // we can make it use less RAM. With other filters we don't 206 // know what to do. 207 size_t i = 0; 208 while (filters[i].id != LZMA_FILTER_LZMA2 209 && filters[i].id != LZMA_FILTER_LZMA1) { 210 if (filters[i].id == LZMA_VLI_UNKNOWN) 211 memlimit_too_small(memory_usage); 212 213 ++i; 214 } 215 216 // Decrease the dictionary size until we meet the memory 217 // usage limit. First round down to full mebibytes. 218 lzma_options_lzma *opt = filters[i].options; 219 const uint32_t orig_dict_size = opt->dict_size; 220 opt->dict_size &= ~((UINT32_C(1) << 20) - 1); 221 while (true) { 222 // If it is below 1 MiB, auto-adjusting failed. We 223 // could be more sophisticated and scale it down even 224 // more, but let's see if many complain about this 225 // version. 226 // 227 // FIXME: Displays the scaled memory usage instead 228 // of the original. 229 if (opt->dict_size < (UINT32_C(1) << 20)) 230 memlimit_too_small(memory_usage); 231 232 memory_usage = lzma_raw_encoder_memusage(filters); 233 if (memory_usage == UINT64_MAX) 234 message_bug(); 235 236 // Accept it if it is low enough. 237 if (memory_usage <= memory_limit) 238 break; 239 240 // Otherwise 1 MiB down and try again. I hope this 241 // isn't too slow method for cases where the original 242 // dict_size is very big. 243 opt->dict_size -= UINT32_C(1) << 20; 244 } 245 246 // Tell the user that we decreased the dictionary size. 247 message(V_WARNING, _("Adjusted LZMA%c dictionary size " 248 "from %s MiB to %s MiB to not exceed " 249 "the memory usage limit of %s MiB"), 250 filters[i].id == LZMA_FILTER_LZMA2 251 ? '2' : '1', 252 uint64_to_str(orig_dict_size >> 20, 0), 253 uint64_to_str(opt->dict_size >> 20, 1), 254 uint64_to_str(round_up_to_mib( 255 memory_limit), 2)); 256 } 257 258 /* 259 // Limit the number of worker threads so that memory usage 260 // limit isn't exceeded. 261 assert(memory_usage > 0); 262 size_t thread_limit = memory_limit / memory_usage; 263 if (thread_limit == 0) 264 thread_limit = 1; 265 266 if (opt_threads > thread_limit) 267 opt_threads = thread_limit; 268 */ 269 270 if (check_default) { 271 // The default check type is CRC64, but fallback to CRC32 272 // if CRC64 isn't supported by the copy of liblzma we are 273 // using. CRC32 is always supported. 274 check = LZMA_CHECK_CRC64; 275 if (!lzma_check_is_supported(check)) 276 check = LZMA_CHECK_CRC32; 277 } 278 279 return; 280 } 281 282 283 /// Return true if the data in in_buf seems to be in the .xz format. 284 static bool 285 is_format_xz(void) 286 { 287 // Specify the magic as hex to be compatible with EBCDIC systems. 288 static const uint8_t magic[6] = { 0xFD, 0x37, 0x7A, 0x58, 0x5A, 0x00 }; 289 return strm.avail_in >= sizeof(magic) 290 && memcmp(in_buf.u8, magic, sizeof(magic)) == 0; 291 } 292 293 294 /// Return true if the data in in_buf seems to be in the .lzma format. 295 static bool 296 is_format_lzma(void) 297 { 298 // The .lzma header is 13 bytes. 299 if (strm.avail_in < 13) 300 return false; 301 302 // Decode the LZMA1 properties. 303 lzma_filter filter = { .id = LZMA_FILTER_LZMA1 }; 304 if (lzma_properties_decode(&filter, NULL, in_buf.u8, 5) != LZMA_OK) 305 return false; 306 307 // A hack to ditch tons of false positives: We allow only dictionary 308 // sizes that are 2^n or 2^n + 2^(n-1) or UINT32_MAX. LZMA_Alone 309 // created only files with 2^n, but accepts any dictionary size. 310 // If someone complains, this will be reconsidered. 311 lzma_options_lzma *opt = filter.options; 312 const uint32_t dict_size = opt->dict_size; 313 free(opt); 314 315 if (dict_size != UINT32_MAX) { 316 uint32_t d = dict_size - 1; 317 d |= d >> 2; 318 d |= d >> 3; 319 d |= d >> 4; 320 d |= d >> 8; 321 d |= d >> 16; 322 ++d; 323 if (d != dict_size || dict_size == 0) 324 return false; 325 } 326 327 // Another hack to ditch false positives: Assume that if the 328 // uncompressed size is known, it must be less than 256 GiB. 329 // Again, if someone complains, this will be reconsidered. 330 uint64_t uncompressed_size = 0; 331 for (size_t i = 0; i < 8; ++i) 332 uncompressed_size |= (uint64_t)(in_buf.u8[5 + i]) << (i * 8); 333 334 if (uncompressed_size != UINT64_MAX 335 && uncompressed_size > (UINT64_C(1) << 38)) 336 return false; 337 338 return true; 339 } 340 341 342 /// Detect the input file type (for now, this done only when decompressing), 343 /// and initialize an appropriate coder. Return value indicates if a normal 344 /// liblzma-based coder was initialized (CODER_INIT_NORMAL), if passthru 345 /// mode should be used (CODER_INIT_PASSTHRU), or if an error occurred 346 /// (CODER_INIT_ERROR). 347 static enum coder_init_ret 348 coder_init(file_pair *pair) 349 { 350 lzma_ret ret = LZMA_PROG_ERROR; 351 352 if (opt_mode == MODE_COMPRESS) { 353 switch (opt_format) { 354 case FORMAT_AUTO: 355 // args.c ensures this. 356 assert(0); 357 break; 358 359 case FORMAT_XZ: 360 ret = lzma_stream_encoder(&strm, filters, check); 361 break; 362 363 case FORMAT_LZMA: 364 ret = lzma_alone_encoder(&strm, filters[0].options); 365 break; 366 367 case FORMAT_RAW: 368 ret = lzma_raw_encoder(&strm, filters); 369 break; 370 } 371 } else { 372 const uint32_t flags = LZMA_TELL_UNSUPPORTED_CHECK 373 | LZMA_CONCATENATED; 374 375 // We abuse FORMAT_AUTO to indicate unknown file format, 376 // for which we may consider passthru mode. 377 enum format_type init_format = FORMAT_AUTO; 378 379 switch (opt_format) { 380 case FORMAT_AUTO: 381 if (is_format_xz()) 382 init_format = FORMAT_XZ; 383 else if (is_format_lzma()) 384 init_format = FORMAT_LZMA; 385 break; 386 387 case FORMAT_XZ: 388 if (is_format_xz()) 389 init_format = FORMAT_XZ; 390 break; 391 392 case FORMAT_LZMA: 393 if (is_format_lzma()) 394 init_format = FORMAT_LZMA; 395 break; 396 397 case FORMAT_RAW: 398 init_format = FORMAT_RAW; 399 break; 400 } 401 402 switch (init_format) { 403 case FORMAT_AUTO: 404 // Uknown file format. If --decompress --stdout 405 // --force have been given, then we copy the input 406 // as is to stdout. Checking for MODE_DECOMPRESS 407 // is needed, because we don't want to do use 408 // passthru mode with --test. 409 if (opt_mode == MODE_DECOMPRESS 410 && opt_stdout && opt_force) 411 return CODER_INIT_PASSTHRU; 412 413 ret = LZMA_FORMAT_ERROR; 414 break; 415 416 case FORMAT_XZ: 417 ret = lzma_stream_decoder(&strm, 418 hardware_memlimit_get( 419 MODE_DECOMPRESS), flags); 420 break; 421 422 case FORMAT_LZMA: 423 ret = lzma_alone_decoder(&strm, 424 hardware_memlimit_get( 425 MODE_DECOMPRESS)); 426 break; 427 428 case FORMAT_RAW: 429 // Memory usage has already been checked in 430 // coder_set_compression_settings(). 431 ret = lzma_raw_decoder(&strm, filters); 432 break; 433 } 434 435 // Try to decode the headers. This will catch too low 436 // memory usage limit in case it happens in the first 437 // Block of the first Stream, which is where it very 438 // probably will happen if it is going to happen. 439 if (ret == LZMA_OK && init_format != FORMAT_RAW) { 440 strm.next_out = NULL; 441 strm.avail_out = 0; 442 ret = lzma_code(&strm, LZMA_RUN); 443 } 444 } 445 446 if (ret != LZMA_OK) { 447 message_error("%s: %s", pair->src_name, message_strm(ret)); 448 if (ret == LZMA_MEMLIMIT_ERROR) 449 message_mem_needed(V_ERROR, lzma_memusage(&strm)); 450 451 return CODER_INIT_ERROR; 452 } 453 454 return CODER_INIT_NORMAL; 455 } 456 457 458 /// Compress or decompress using liblzma. 459 static bool 460 coder_normal(file_pair *pair) 461 { 462 // Encoder needs to know when we have given all the input to it. 463 // The decoders need to know it too when we are using 464 // LZMA_CONCATENATED. We need to check for src_eof here, because 465 // the first input chunk has been already read, and that may 466 // have been the only chunk we will read. 467 lzma_action action = pair->src_eof ? LZMA_FINISH : LZMA_RUN; 468 469 lzma_ret ret; 470 471 // Assume that something goes wrong. 472 bool success = false; 473 474 strm.next_out = out_buf.u8; 475 strm.avail_out = IO_BUFFER_SIZE; 476 477 while (!user_abort) { 478 // Fill the input buffer if it is empty and we haven't reached 479 // end of file yet. 480 if (strm.avail_in == 0 && !pair->src_eof) { 481 strm.next_in = in_buf.u8; 482 strm.avail_in = io_read( 483 pair, &in_buf, IO_BUFFER_SIZE); 484 485 if (strm.avail_in == SIZE_MAX) 486 break; 487 488 if (pair->src_eof) 489 action = LZMA_FINISH; 490 } 491 492 // Let liblzma do the actual work. 493 ret = lzma_code(&strm, action); 494 495 // Write out if the output buffer became full. 496 if (strm.avail_out == 0) { 497 if (opt_mode != MODE_TEST && io_write(pair, &out_buf, 498 IO_BUFFER_SIZE - strm.avail_out)) 499 break; 500 501 strm.next_out = out_buf.u8; 502 strm.avail_out = IO_BUFFER_SIZE; 503 } 504 505 if (ret != LZMA_OK) { 506 // Determine if the return value indicates that we 507 // won't continue coding. 508 const bool stop = ret != LZMA_NO_CHECK 509 && ret != LZMA_UNSUPPORTED_CHECK; 510 511 if (stop) { 512 // Write the remaining bytes even if something 513 // went wrong, because that way the user gets 514 // as much data as possible, which can be good 515 // when trying to get at least some useful 516 // data out of damaged files. 517 if (opt_mode != MODE_TEST && io_write(pair, 518 &out_buf, IO_BUFFER_SIZE 519 - strm.avail_out)) 520 break; 521 } 522 523 if (ret == LZMA_STREAM_END) { 524 // Check that there is no trailing garbage. 525 // This is needed for LZMA_Alone and raw 526 // streams. 527 if (strm.avail_in == 0 && !pair->src_eof) { 528 // Try reading one more byte. 529 // Hopefully we don't get any more 530 // input, and thus pair->src_eof 531 // becomes true. 532 strm.avail_in = io_read( 533 pair, &in_buf, 1); 534 if (strm.avail_in == SIZE_MAX) 535 break; 536 537 assert(strm.avail_in == 0 538 || strm.avail_in == 1); 539 } 540 541 if (strm.avail_in == 0) { 542 assert(pair->src_eof); 543 success = true; 544 break; 545 } 546 547 // We hadn't reached the end of the file. 548 ret = LZMA_DATA_ERROR; 549 assert(stop); 550 } 551 552 // If we get here and stop is true, something went 553 // wrong and we print an error. Otherwise it's just 554 // a warning and coding can continue. 555 if (stop) { 556 message_error("%s: %s", pair->src_name, 557 message_strm(ret)); 558 } else { 559 message_warning("%s: %s", pair->src_name, 560 message_strm(ret)); 561 562 // When compressing, all possible errors set 563 // stop to true. 564 assert(opt_mode != MODE_COMPRESS); 565 } 566 567 if (ret == LZMA_MEMLIMIT_ERROR) { 568 // Display how much memory it would have 569 // actually needed. 570 message_mem_needed(V_ERROR, 571 lzma_memusage(&strm)); 572 } 573 574 if (stop) 575 break; 576 } 577 578 // Show progress information under certain conditions. 579 message_progress_update(); 580 } 581 582 return success; 583 } 584 585 586 /// Copy from input file to output file without processing the data in any 587 /// way. This is used only when trying to decompress unrecognized files 588 /// with --decompress --stdout --force, so the output is always stdout. 589 static bool 590 coder_passthru(file_pair *pair) 591 { 592 while (strm.avail_in != 0) { 593 if (user_abort) 594 return false; 595 596 if (io_write(pair, &in_buf, strm.avail_in)) 597 return false; 598 599 strm.total_in += strm.avail_in; 600 strm.total_out = strm.total_in; 601 message_progress_update(); 602 603 strm.avail_in = io_read(pair, &in_buf, IO_BUFFER_SIZE); 604 if (strm.avail_in == SIZE_MAX) 605 return false; 606 } 607 608 return true; 609 } 610 611 612 extern void 613 coder_run(const char *filename) 614 { 615 // Set and possibly print the filename for the progress message. 616 message_filename(filename); 617 618 // Try to open the input file. 619 file_pair *pair = io_open_src(filename); 620 if (pair == NULL) 621 return; 622 623 // Assume that something goes wrong. 624 bool success = false; 625 626 // Read the first chunk of input data. This is needed to detect 627 // the input file type (for now, only for decompression). 628 strm.next_in = in_buf.u8; 629 strm.avail_in = io_read(pair, &in_buf, IO_BUFFER_SIZE); 630 631 if (strm.avail_in != SIZE_MAX) { 632 // Initialize the coder. This will detect the file format 633 // and, in decompression or testing mode, check the memory 634 // usage of the first Block too. This way we don't try to 635 // open the destination file if we see that coding wouldn't 636 // work at all anyway. This also avoids deleting the old 637 // "target" file if --force was used. 638 const enum coder_init_ret init_ret = coder_init(pair); 639 640 if (init_ret != CODER_INIT_ERROR && !user_abort) { 641 // Don't open the destination file when --test 642 // is used. 643 if (opt_mode == MODE_TEST || !io_open_dest(pair)) { 644 // Initialize the progress indicator. 645 const uint64_t in_size 646 = pair->src_st.st_size <= 0 647 ? 0 : pair->src_st.st_size; 648 message_progress_start(&strm, in_size); 649 650 // Do the actual coding or passthru. 651 if (init_ret == CODER_INIT_NORMAL) 652 success = coder_normal(pair); 653 else 654 success = coder_passthru(pair); 655 656 message_progress_end(success); 657 } 658 } 659 } 660 661 // Close the file pair. It needs to know if coding was successful to 662 // know if the source or target file should be unlinked. 663 io_close(pair, success); 664 665 return; 666 } 667