src/xz/process.c

   1 ///////////////////////////////////////////////////////////////////////////////
   2 //
   3 /// \file       process.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 enum operation_mode opt_mode = MODE_COMPRESS;
  17
  18 enum format_type opt_format = FORMAT_AUTO;
  19
  20
  21 /// Stream used to communicate with liblzma
  22 static lzma_stream strm = LZMA_STREAM_INIT;
  23
  24 /// Filters needed for all encoding all formats, and also decoding in raw data
  25 static lzma_filter filters[LZMA_FILTERS_MAX + 1];
  26
  27 /// Number of filters. Zero indicates that we are using a preset.
  28 static size_t filters_count = 0;
  29
  30 /// Number of the preset (0-9)
  31 static size_t preset_number = 6;
  32
  33 /// True if we should auto-adjust the compression settings to use less memory
  34 /// if memory usage limit is too low for the original settings.
  35 static bool auto_adjust = true;
  36
  37 /// Indicate if no preset has been explicitly given. In that case, if we need
  38 /// to auto-adjust for lower memory usage, we won't print a warning.
  39 static bool preset_default = true;
  40
  41 /// If a preset is used (no custom filter chain) and preset_extreme is true,
  42 /// a significantly slower compression is used to achieve slightly better
  43 /// compression ratio.
  44 static bool preset_extreme = false;
  45
  46 /// Integrity check type
  47 #ifdef HAVE_CHECK_CRC64
  48 static lzma_check check = LZMA_CHECK_CRC64;
  49 #else
  50 static lzma_check check = LZMA_CHECK_CRC32;
  51 #endif
  52
  53
  54 extern void
  55 coder_set_check(lzma_check new_check)
  56 {
  57         check = new_check;
  58         return;
  59 }
  60
  61
  62 extern void
  63 coder_set_preset(size_t new_preset)
  64 {
  65         preset_number = new_preset;
  66         preset_default = false;
  67         return;
  68 }
  69
  70
  71 extern void
  72 coder_set_extreme(void)
  73 {
  74         preset_extreme = true;
  75         return;
  76 }
  77
  78
  79 extern void
  80 coder_add_filter(lzma_vli id, void *options)
  81 {
  82         if (filters_count == LZMA_FILTERS_MAX)
  83                 message_fatal(_("Maximum number of filters is four"));
  84
  85         filters[filters_count].id = id;
  86         filters[filters_count].options = options;
  87         ++filters_count;
  88
  89         return;
  90 }
  91
  92
  93 static void lzma_attribute((noreturn))
  94 memlimit_too_small(uint64_t memory_usage, uint64_t memory_limit)
  95 {
  96         message_fatal(_("Memory usage limit (%" PRIu64 " MiB) is too small "
  97                         "for the given filter setup (%" PRIu64 " MiB)"),
  98                         memory_limit >> 20, memory_usage >> 20);
  99 }
 100
 101
 102 extern void
 103 coder_set_compression_settings(void)
 104 {
 105         // Options for LZMA1 or LZMA2 in case we are using a preset.
 106         static lzma_options_lzma opt_lzma;
 107
 108         if (filters_count == 0) {
 109                 // We are using a preset. This is not a good idea in raw mode
 110                 // except when playing around with things. Different versions
 111                 // of this software may use different options in presets, and
 112                 // thus make uncompressing the raw data difficult.
 113                 if (opt_format == FORMAT_RAW) {
 114                         // The message is shown only if warnings are allowed
 115                         // but the exit status isn't changed.
 116                         message(V_WARNING, _("Using a preset in raw mode "
 117                                         "is discouraged."));
 118                         message(V_WARNING, _("The exact options of the "
 119                                         "presets may vary between software "
 120                                         "versions."));
 121                 }
 122
 123                 // Get the preset for LZMA1 or LZMA2.
 124                 if (preset_extreme)
 125                         preset_number |= LZMA_PRESET_EXTREME;
 126
 127                 if (lzma_lzma_preset(&opt_lzma, preset_number))
 128                         message_bug();
 129
 130                 // Use LZMA2 except with --format=lzma we use LZMA1.
 131                 filters[0].id = opt_format == FORMAT_LZMA
 132                                 ? LZMA_FILTER_LZMA1 : LZMA_FILTER_LZMA2;
 133                 filters[0].options = &opt_lzma;
 134                 filters_count = 1;
 135         } else {
 136                 preset_default = false;
 137         }
 138
 139         // Terminate the filter options array.
 140         filters[filters_count].id = LZMA_VLI_UNKNOWN;
 141
 142         // If we are using the LZMA_Alone format, allow exactly one filter
 143         // which has to be LZMA.
 144         if (opt_format == FORMAT_LZMA && (filters_count != 1
 145                         || filters[0].id != LZMA_FILTER_LZMA1))
 146                 message_fatal(_("With --format=lzma only the LZMA1 filter "
 147                                 "is supported"));
 148
 149         // Print the selected filter chain.
 150         message_filters(V_DEBUG, filters);
 151
 152         // If using --format=raw, we can be decoding. The memusage function
 153         // also validates the filter chain and the options used for the
 154         // filters.
 155         uint64_t memory_usage;
 156         uint64_t memory_limit;
 157         if (opt_mode == MODE_COMPRESS) {
 158                 memory_usage = lzma_raw_encoder_memusage(filters);
 159                 memory_limit = hardware_memlimit_encoder();
 160         } else {
 161                 memory_usage = lzma_raw_decoder_memusage(filters);
 162                 memory_limit = hardware_memlimit_decoder();
 163         }
 164
 165         if (memory_usage == UINT64_MAX)
 166                 message_fatal("Unsupported filter chain or filter options");
 167
 168         // Print memory usage info.
 169         message(V_DEBUG, _("%'" PRIu64 " MiB (%'" PRIu64 " B) of memory is "
 170                         "required per thread, "
 171                         "limit is %'" PRIu64 " MiB (%'" PRIu64 " B)"),
 172                         memory_usage >> 20, memory_usage,
 173                         memory_limit >> 20, memory_limit);
 174
 175         if (memory_usage > memory_limit) {
 176                 // If --no-auto-adjust was used or we didn't find LZMA1 or
 177                 // LZMA2 as the last filter, give an error immediatelly.
 178                 // --format=raw implies --no-auto-adjust.
 179                 if (!auto_adjust || opt_format == FORMAT_RAW)
 180                         memlimit_too_small(memory_usage, memory_limit);
 181
 182                 assert(opt_mode == MODE_COMPRESS);
 183
 184                 // Look for the last filter if it is LZMA2 or LZMA1, so
 185                 // we can make it use less RAM. With other filters we don't
 186                 // know what to do.
 187                 size_t i = 0;
 188                 while (filters[i].id != LZMA_FILTER_LZMA2
 189                                 && filters[i].id != LZMA_FILTER_LZMA1) {
 190                         if (filters[i].id == LZMA_VLI_UNKNOWN)
 191                                 memlimit_too_small(memory_usage, memory_limit);
 192
 193                         ++i;
 194                 }
 195
 196                 // Decrease the dictionary size until we meet the memory
 197                 // usage limit. First round down to full mebibytes.
 198                 lzma_options_lzma *opt = filters[i].options;
 199                 const uint32_t orig_dict_size = opt->dict_size;
 200                 opt->dict_size &= ~((UINT32_C(1) << 20) - 1);
 201                 while (true) {
 202                         // If it is below 1 MiB, auto-adjusting failed. We
 203                         // could be more sophisticated and scale it down even
 204                         // more, but let's see if many complain about this
 205                         // version.
 206                         //
 207                         // FIXME: Displays the scaled memory usage instead
 208                         // of the original.
 209                         if (opt->dict_size < (UINT32_C(1) << 20))
 210                                 memlimit_too_small(memory_usage, memory_limit);
 211
 212                         memory_usage = lzma_raw_encoder_memusage(filters);
 213                         if (memory_usage == UINT64_MAX)
 214                                 message_bug();
 215
 216                         // Accept it if it is low enough.
 217                         if (memory_usage <= memory_limit)
 218                                 break;
 219
 220                         // Otherwise 1 MiB down and try again. I hope this
 221                         // isn't too slow method for cases where the original
 222                         // dict_size is very big.
 223                         opt->dict_size -= UINT32_C(1) << 20;
 224                 }
 225
 226                 // Tell the user that we decreased the dictionary size.
 227                 // However, omit the message if no preset or custom chain
 228                 // was given. FIXME: Always warn?
 229                 if (!preset_default)
 230                         message(V_WARNING, "Adjusted LZMA%c dictionary size "
 231                                         "from %'" PRIu32 " MiB to "
 232                                         "%'" PRIu32 " MiB to not exceed "
 233                                         "the memory usage limit of "
 234                                         "%'" PRIu64 " MiB",
 235                                         filters[i].id == LZMA_FILTER_LZMA2
 236                                                 ? '2' : '1',
 237                                         orig_dict_size >> 20,
 238                                         opt->dict_size >> 20,
 239                                         memory_limit >> 20);
 240         }
 241
 242 /*
 243         // Limit the number of worker threads so that memory usage
 244         // limit isn't exceeded.
 245         assert(memory_usage > 0);
 246         size_t thread_limit = memory_limit / memory_usage;
 247         if (thread_limit == 0)
 248                 thread_limit = 1;
 249
 250         if (opt_threads > thread_limit)
 251                 opt_threads = thread_limit;
 252 */
 253
 254         return;
 255 }
 256
 257
 258 static bool
 259 coder_init(void)
 260 {
 261         lzma_ret ret = LZMA_PROG_ERROR;
 262
 263         if (opt_mode == MODE_COMPRESS) {
 264                 switch (opt_format) {
 265                 case FORMAT_AUTO:
 266                         // args.c ensures this.
 267                         assert(0);
 268                         break;
 269
 270                 case FORMAT_XZ:
 271                         ret = lzma_stream_encoder(&strm, filters, check);
 272                         break;
 273
 274                 case FORMAT_LZMA:
 275                         ret = lzma_alone_encoder(&strm, filters[0].options);
 276                         break;
 277
 278                 case FORMAT_RAW:
 279                         ret = lzma_raw_encoder(&strm, filters);
 280                         break;
 281                 }
 282         } else {
 283                 const uint32_t flags = LZMA_TELL_UNSUPPORTED_CHECK
 284                                 | LZMA_CONCATENATED;
 285
 286                 switch (opt_format) {
 287                 case FORMAT_AUTO:
 288                         ret = lzma_auto_decoder(&strm,
 289                                         hardware_memlimit_decoder(), flags);
 290                         break;
 291
 292                 case FORMAT_XZ:
 293                         ret = lzma_stream_decoder(&strm,
 294                                         hardware_memlimit_decoder(), flags);
 295                         break;
 296
 297                 case FORMAT_LZMA:
 298                         ret = lzma_alone_decoder(&strm,
 299                                         hardware_memlimit_decoder());
 300                         break;
 301
 302                 case FORMAT_RAW:
 303                         // Memory usage has already been checked in
 304                         // coder_set_compression_settings().
 305                         ret = lzma_raw_decoder(&strm, filters);
 306                         break;
 307                 }
 308         }
 309
 310         if (ret != LZMA_OK) {
 311                 if (ret == LZMA_MEM_ERROR)
 312                         message_error("%s", message_strm(LZMA_MEM_ERROR));
 313                 else
 314                         message_bug();
 315
 316                 return true;
 317         }
 318
 319         return false;
 320 }
 321
 322
 323 static bool
 324 coder_run(file_pair *pair)
 325 {
 326         // Buffers to hold input and output data.
 327         uint8_t in_buf[IO_BUFFER_SIZE];
 328         uint8_t out_buf[IO_BUFFER_SIZE];
 329
 330         // Initialize the progress indicator.
 331         const uint64_t in_size = pair->src_st.st_size <= (off_t)(0)
 332                         ? 0 : (uint64_t)(pair->src_st.st_size);
 333         message_progress_start(&strm, pair->src_name, in_size);
 334
 335         lzma_action action = LZMA_RUN;
 336         lzma_ret ret;
 337         bool success = false; // Assume that something goes wrong.
 338
 339         strm.avail_in = 0;
 340         strm.next_out = out_buf;
 341         strm.avail_out = IO_BUFFER_SIZE;
 342
 343         while (!user_abort) {
 344                 // Fill the input buffer if it is empty and we haven't reached
 345                 // end of file yet.
 346                 if (strm.avail_in == 0 && !pair->src_eof) {
 347                         strm.next_in = in_buf;
 348                         strm.avail_in = io_read(pair, in_buf, IO_BUFFER_SIZE);
 349
 350                         if (strm.avail_in == SIZE_MAX)
 351                                 break;
 352
 353                         // Encoder needs to know when we have given all the
 354                         // input to it. The decoders need to know it too when
 355                         // we are using LZMA_CONCATENATED.
 356                         if (pair->src_eof)
 357                                 action = LZMA_FINISH;
 358                 }
 359
 360                 // Let liblzma do the actual work.
 361                 ret = lzma_code(&strm, action);
 362
 363                 // Write out if the output buffer became full.
 364                 if (strm.avail_out == 0) {
 365                         if (opt_mode != MODE_TEST && io_write(pair, out_buf,
 366                                         IO_BUFFER_SIZE - strm.avail_out))
 367                                 break;
 368
 369                         strm.next_out = out_buf;
 370                         strm.avail_out = IO_BUFFER_SIZE;
 371                 }
 372
 373                 if (ret != LZMA_OK) {
 374                         // Determine if the return value indicates that we
 375                         // won't continue coding.
 376                         const bool stop = ret != LZMA_NO_CHECK
 377                                         && ret != LZMA_UNSUPPORTED_CHECK;
 378
 379                         if (stop) {
 380                                 // Write the remaining bytes even if something
 381                                 // went wrong, because that way the user gets
 382                                 // as much data as possible, which can be good
 383                                 // when trying to get at least some useful
 384                                 // data out of damaged files.
 385                                 if (opt_mode != MODE_TEST && io_write(pair,
 386                                                 out_buf, IO_BUFFER_SIZE
 387                                                         - strm.avail_out))
 388                                         break;
 389                         }
 390
 391                         if (ret == LZMA_STREAM_END) {
 392                                 // Check that there is no trailing garbage.
 393                                 // This is needed for LZMA_Alone and raw
 394                                 // streams.
 395                                 if (strm.avail_in == 0 && !pair->src_eof) {
 396                                         // Try reading one more byte.
 397                                         // Hopefully we don't get any more
 398                                         // input, and thus pair->src_eof
 399                                         // becomes true.
 400                                         strm.avail_in = io_read(
 401                                                         pair, in_buf, 1);
 402                                         if (strm.avail_in == SIZE_MAX)
 403                                                 break;
 404
 405                                         assert(strm.avail_in == 0
 406                                                         || strm.avail_in == 1);
 407                                 }
 408
 409                                 if (strm.avail_in == 0) {
 410                                         assert(pair->src_eof);
 411                                         success = true;
 412                                         break;
 413                                 }
 414
 415                                 // We hadn't reached the end of the file.
 416                                 ret = LZMA_DATA_ERROR;
 417                                 assert(stop);
 418                         }
 419
 420                         // If we get here and stop is true, something went
 421                         // wrong and we print an error. Otherwise it's just
 422                         // a warning and coding can continue.
 423                         if (stop) {
 424                                 message_error("%s: %s", pair->src_name,
 425                                                 message_strm(ret));
 426                         } else {
 427                                 message_warning("%s: %s", pair->src_name,
 428                                                 message_strm(ret));
 429
 430                                 // When compressing, all possible errors set
 431                                 // stop to true.
 432                                 assert(opt_mode != MODE_COMPRESS);
 433                         }
 434
 435                         if (ret == LZMA_MEMLIMIT_ERROR) {
 436                                 // Figure out how much memory it would have
 437                                 // actually needed.
 438                                 uint64_t memusage = lzma_memusage(&strm);
 439                                 uint64_t memlimit
 440                                                 = hardware_memlimit_decoder();
 441
 442                                 // Round the memory limit down and usage up.
 443                                 // This way we don't display a ridiculous
 444                                 // message like "Limit was 9 MiB, but 9 MiB
 445                                 // would have been needed".
 446                                 memusage = (memusage + 1024 * 1024 - 1)
 447                                                 / (1024 * 1024);
 448                                 memlimit /= 1024 * 1024;
 449
 450                                 message_error(_("Limit was %'" PRIu64 " MiB, "
 451                                                 "but %'" PRIu64 " MiB would "
 452                                                 "have been needed"),
 453                                                 memlimit, memusage);
 454                         }
 455
 456                         if (stop)
 457                                 break;
 458                 }
 459
 460                 // Show progress information under certain conditions.
 461                 message_progress_update();
 462         }
 463
 464         message_progress_end(success);
 465
 466         return success;
 467 }
 468
 469
 470 extern void
 471 process_file(const char *filename)
 472 {
 473         // First try initializing the coder. If it fails, it's useless to try
 474         // opening the file. Check also for user_abort just in case if we had
 475         // got a signal while initializing the coder.
 476         if (coder_init() || user_abort)
 477                 return;
 478
 479         // Try to open the input and output files.
 480         file_pair *pair = io_open(filename);
 481         if (pair == NULL)
 482                 return;
 483
 484         // Do the actual coding.
 485         const bool success = coder_run(pair);
 486
 487         // Close the file pair. It needs to know if coding was successful to
 488         // know if the source or target file should be unlinked.
 489         io_close(pair, success);
 490
 491         return;
 492 }