1 ///////////////////////////////////////////////////////////////////////////////
4 /// \brief Compresses or uncompresses a file
6 // Author: Lasse Collin
8 // This file has been put into the public domain.
9 // You can do whatever you want with this file.
11 ///////////////////////////////////////////////////////////////////////////////
16 /// Return value type for coder_init().
24 enum operation_mode opt_mode = MODE_COMPRESS;
25 enum format_type opt_format = FORMAT_AUTO;
26 bool opt_auto_adjust = true;
29 /// Stream used to communicate with liblzma
30 static lzma_stream strm = LZMA_STREAM_INIT;
32 /// Filters needed for all encoding all formats, and also decoding in raw data
33 static lzma_filter filters[LZMA_FILTERS_MAX + 1];
35 /// Input and output buffers
37 static io_buf out_buf;
39 /// Number of filters. Zero indicates that we are using a preset.
40 static size_t filters_count = 0;
42 /// Number of the preset (0-9)
43 static size_t preset_number = 6;
45 /// Indicate if no preset has been explicitly given. In that case, if we need
46 /// to auto-adjust for lower memory usage, we won't print a warning.
47 static bool preset_default = true;
49 /// If a preset is used (no custom filter chain) and preset_extreme is true,
50 /// a significantly slower compression is used to achieve slightly better
51 /// compression ratio.
52 static bool preset_extreme = false;
54 /// Integrity check type
55 static lzma_check check;
57 /// This becomes false if the --check=CHECK option is used.
58 static bool check_default = true;
62 coder_set_check(lzma_check new_check)
65 check_default = false;
71 coder_set_preset(size_t new_preset)
73 preset_number = new_preset;
74 preset_default = false;
80 coder_set_extreme(void)
82 preset_extreme = true;
88 coder_add_filter(lzma_vli id, void *options)
90 if (filters_count == LZMA_FILTERS_MAX)
91 message_fatal(_("Maximum number of filters is four"));
93 filters[filters_count].id = id;
94 filters[filters_count].options = options;
101 static void lzma_attribute((noreturn))
102 memlimit_too_small(uint64_t memory_usage)
104 message(V_ERROR, _("Memory usage limit is too low for the given "
106 message_mem_needed(V_ERROR, memory_usage);
107 tuklib_exit(E_ERROR, E_ERROR, false);
112 coder_set_compression_settings(void)
114 // Options for LZMA1 or LZMA2 in case we are using a preset.
115 static lzma_options_lzma opt_lzma;
117 if (filters_count == 0) {
118 // We are using a preset. This is not a good idea in raw mode
119 // except when playing around with things. Different versions
120 // of this software may use different options in presets, and
121 // thus make uncompressing the raw data difficult.
122 if (opt_format == FORMAT_RAW) {
123 // The message is shown only if warnings are allowed
124 // but the exit status isn't changed.
125 message(V_WARNING, _("Using a preset in raw mode "
127 message(V_WARNING, _("The exact options of the "
128 "presets may vary between software "
132 // Get the preset for LZMA1 or LZMA2.
134 preset_number |= LZMA_PRESET_EXTREME;
136 if (lzma_lzma_preset(&opt_lzma, preset_number))
139 // Use LZMA2 except with --format=lzma we use LZMA1.
140 filters[0].id = opt_format == FORMAT_LZMA
141 ? LZMA_FILTER_LZMA1 : LZMA_FILTER_LZMA2;
142 filters[0].options = &opt_lzma;
145 preset_default = false;
148 // Terminate the filter options array.
149 filters[filters_count].id = LZMA_VLI_UNKNOWN;
151 // If we are using the .lzma format, allow exactly one filter
152 // which has to be LZMA1.
153 if (opt_format == FORMAT_LZMA && (filters_count != 1
154 || filters[0].id != LZMA_FILTER_LZMA1))
155 message_fatal(_("The .lzma format supports only "
156 "the LZMA1 filter"));
158 // If we are using the .xz format, make sure that there is no LZMA1
159 // filter to prevent LZMA_PROG_ERROR.
160 if (opt_format == FORMAT_XZ)
161 for (size_t i = 0; i < filters_count; ++i)
162 if (filters[i].id == LZMA_FILTER_LZMA1)
163 message_fatal(_("LZMA1 cannot be used "
164 "with the .xz format"));
166 // Print the selected filter chain.
167 message_filters_show(V_DEBUG, filters);
169 // If using --format=raw, we can be decoding. The memusage function
170 // also validates the filter chain and the options used for the
172 const uint64_t memory_limit = hardware_memlimit_get(opt_mode);
173 uint64_t memory_usage;
174 if (opt_mode == MODE_COMPRESS)
175 memory_usage = lzma_raw_encoder_memusage(filters);
177 memory_usage = lzma_raw_decoder_memusage(filters);
179 if (memory_usage == UINT64_MAX)
180 message_fatal(_("Unsupported filter chain or filter options"));
182 // Print memory usage info before possible dictionary
183 // size auto-adjusting.
184 message_mem_needed(V_DEBUG, memory_usage);
186 if (memory_usage > memory_limit) {
187 // If --no-auto-adjust was used or we didn't find LZMA1 or
188 // LZMA2 as the last filter, give an error immediately.
189 // --format=raw implies --no-auto-adjust.
190 if (!opt_auto_adjust || opt_format == FORMAT_RAW)
191 memlimit_too_small(memory_usage);
193 assert(opt_mode == MODE_COMPRESS);
195 // Look for the last filter if it is LZMA2 or LZMA1, so
196 // we can make it use less RAM. With other filters we don't
199 while (filters[i].id != LZMA_FILTER_LZMA2
200 && filters[i].id != LZMA_FILTER_LZMA1) {
201 if (filters[i].id == LZMA_VLI_UNKNOWN)
202 memlimit_too_small(memory_usage);
207 // Decrease the dictionary size until we meet the memory
208 // usage limit. First round down to full mebibytes.
209 lzma_options_lzma *opt = filters[i].options;
210 const uint32_t orig_dict_size = opt->dict_size;
211 opt->dict_size &= ~((UINT32_C(1) << 20) - 1);
213 // If it is below 1 MiB, auto-adjusting failed. We
214 // could be more sophisticated and scale it down even
215 // more, but let's see if many complain about this
218 // FIXME: Displays the scaled memory usage instead
220 if (opt->dict_size < (UINT32_C(1) << 20))
221 memlimit_too_small(memory_usage);
223 memory_usage = lzma_raw_encoder_memusage(filters);
224 if (memory_usage == UINT64_MAX)
227 // Accept it if it is low enough.
228 if (memory_usage <= memory_limit)
231 // Otherwise 1 MiB down and try again. I hope this
232 // isn't too slow method for cases where the original
233 // dict_size is very big.
234 opt->dict_size -= UINT32_C(1) << 20;
237 // Tell the user that we decreased the dictionary size.
238 // However, omit the message if no preset or custom chain
239 // was given. FIXME: Always warn?
241 message(V_WARNING, _("Adjusted LZMA%c dictionary size "
242 "from %s MiB to %s MiB to not exceed "
243 "the memory usage limit of %s MiB"),
244 filters[i].id == LZMA_FILTER_LZMA2
246 uint64_to_str(orig_dict_size >> 20, 0),
247 uint64_to_str(opt->dict_size >> 20, 1),
248 uint64_to_str(round_up_to_mib(
253 // Limit the number of worker threads so that memory usage
254 // limit isn't exceeded.
255 assert(memory_usage > 0);
256 size_t thread_limit = memory_limit / memory_usage;
257 if (thread_limit == 0)
260 if (opt_threads > thread_limit)
261 opt_threads = thread_limit;
265 // The default check type is CRC64, but fallback to CRC32
266 // if CRC64 isn't supported by the copy of liblzma we are
267 // using. CRC32 is always supported.
268 check = LZMA_CHECK_CRC64;
269 if (!lzma_check_is_supported(check))
270 check = LZMA_CHECK_CRC32;
277 /// Return true if the data in in_buf seems to be in the .xz format.
281 return strm.avail_in >= 6 && memcmp(in_buf.u8, "\3757zXZ", 6) == 0;
285 /// Return true if the data in in_buf seems to be in the .lzma format.
289 // The .lzma header is 13 bytes.
290 if (strm.avail_in < 13)
293 // Decode the LZMA1 properties.
294 lzma_filter filter = { .id = LZMA_FILTER_LZMA1 };
295 if (lzma_properties_decode(&filter, NULL, in_buf.u8, 5) != LZMA_OK)
298 // A hack to ditch tons of false positives: We allow only dictionary
299 // sizes that are 2^n or 2^n + 2^(n-1) or UINT32_MAX. LZMA_Alone
300 // created only files with 2^n, but accepts any dictionary size.
301 // If someone complains, this will be reconsidered.
302 lzma_options_lzma *opt = filter.options;
303 const uint32_t dict_size = opt->dict_size;
306 if (dict_size != UINT32_MAX) {
307 uint32_t d = dict_size - 1;
314 if (d != dict_size || dict_size == 0)
318 // Another hack to ditch false positives: Assume that if the
319 // uncompressed size is known, it must be less than 256 GiB.
320 // Again, if someone complains, this will be reconsidered.
321 uint64_t uncompressed_size = 0;
322 for (size_t i = 0; i < 8; ++i)
323 uncompressed_size |= (uint64_t)(in_buf.u8[5 + i]) << (i * 8);
325 if (uncompressed_size != UINT64_MAX
326 && uncompressed_size > (UINT64_C(1) << 38))
333 /// Detect the input file type (for now, this done only when decompressing),
334 /// and initialize an appropriate coder. Return value indicates if a normal
335 /// liblzma-based coder was initialized (CODER_INIT_NORMAL), if passthru
336 /// mode should be used (CODER_INIT_PASSTHRU), or if an error occurred
337 /// (CODER_INIT_ERROR).
338 static enum coder_init_ret
339 coder_init(file_pair *pair)
341 lzma_ret ret = LZMA_PROG_ERROR;
343 if (opt_mode == MODE_COMPRESS) {
344 switch (opt_format) {
346 // args.c ensures this.
351 ret = lzma_stream_encoder(&strm, filters, check);
355 ret = lzma_alone_encoder(&strm, filters[0].options);
359 ret = lzma_raw_encoder(&strm, filters);
363 const uint32_t flags = LZMA_TELL_UNSUPPORTED_CHECK
366 // We abuse FORMAT_AUTO to indicate unknown file format,
367 // for which we may consider passthru mode.
368 enum format_type init_format = FORMAT_AUTO;
370 switch (opt_format) {
373 init_format = FORMAT_XZ;
374 else if (is_format_lzma())
375 init_format = FORMAT_LZMA;
380 init_format = FORMAT_XZ;
384 if (is_format_lzma())
385 init_format = FORMAT_LZMA;
389 init_format = FORMAT_RAW;
393 switch (init_format) {
395 // Uknown file format. If --decompress --stdout
396 // --force have been given, then we copy the input
397 // as is to stdout. Checking for MODE_DECOMPRESS
398 // is needed, because we don't want to do use
399 // passthru mode with --test.
400 if (opt_mode == MODE_DECOMPRESS
401 && opt_stdout && opt_force)
402 return CODER_INIT_PASSTHRU;
404 ret = LZMA_FORMAT_ERROR;
408 ret = lzma_stream_decoder(&strm,
409 hardware_memlimit_get(
410 MODE_DECOMPRESS), flags);
414 ret = lzma_alone_decoder(&strm,
415 hardware_memlimit_get(
420 // Memory usage has already been checked in
421 // coder_set_compression_settings().
422 ret = lzma_raw_decoder(&strm, filters);
426 // Try to decode the headers. This will catch too low
427 // memory usage limit in case it happens in the first
428 // Block of the first Stream, which is where it very
429 // probably will happen if it is going to happen.
430 if (ret == LZMA_OK && init_format != FORMAT_RAW) {
431 strm.next_out = NULL;
433 ret = lzma_code(&strm, LZMA_RUN);
437 if (ret != LZMA_OK) {
438 message_error("%s: %s", pair->src_name, message_strm(ret));
439 if (ret == LZMA_MEMLIMIT_ERROR)
440 message_mem_needed(V_ERROR, lzma_memusage(&strm));
442 return CODER_INIT_ERROR;
445 return CODER_INIT_NORMAL;
449 /// Compress or decompress using liblzma.
451 coder_normal(file_pair *pair)
453 // Encoder needs to know when we have given all the input to it.
454 // The decoders need to know it too when we are using
455 // LZMA_CONCATENATED. We need to check for src_eof here, because
456 // the first input chunk has been already read, and that may
457 // have been the only chunk we will read.
458 lzma_action action = pair->src_eof ? LZMA_FINISH : LZMA_RUN;
462 // Assume that something goes wrong.
463 bool success = false;
465 strm.next_out = out_buf.u8;
466 strm.avail_out = IO_BUFFER_SIZE;
468 while (!user_abort) {
469 // Fill the input buffer if it is empty and we haven't reached
471 if (strm.avail_in == 0 && !pair->src_eof) {
472 strm.next_in = in_buf.u8;
473 strm.avail_in = io_read(
474 pair, &in_buf, IO_BUFFER_SIZE);
476 if (strm.avail_in == SIZE_MAX)
480 action = LZMA_FINISH;
483 // Let liblzma do the actual work.
484 ret = lzma_code(&strm, action);
486 // Write out if the output buffer became full.
487 if (strm.avail_out == 0) {
488 if (opt_mode != MODE_TEST && io_write(pair, &out_buf,
489 IO_BUFFER_SIZE - strm.avail_out))
492 strm.next_out = out_buf.u8;
493 strm.avail_out = IO_BUFFER_SIZE;
496 if (ret != LZMA_OK) {
497 // Determine if the return value indicates that we
498 // won't continue coding.
499 const bool stop = ret != LZMA_NO_CHECK
500 && ret != LZMA_UNSUPPORTED_CHECK;
503 // Write the remaining bytes even if something
504 // went wrong, because that way the user gets
505 // as much data as possible, which can be good
506 // when trying to get at least some useful
507 // data out of damaged files.
508 if (opt_mode != MODE_TEST && io_write(pair,
509 &out_buf, IO_BUFFER_SIZE
514 if (ret == LZMA_STREAM_END) {
515 // Check that there is no trailing garbage.
516 // This is needed for LZMA_Alone and raw
518 if (strm.avail_in == 0 && !pair->src_eof) {
519 // Try reading one more byte.
520 // Hopefully we don't get any more
521 // input, and thus pair->src_eof
523 strm.avail_in = io_read(
525 if (strm.avail_in == SIZE_MAX)
528 assert(strm.avail_in == 0
529 || strm.avail_in == 1);
532 if (strm.avail_in == 0) {
533 assert(pair->src_eof);
538 // We hadn't reached the end of the file.
539 ret = LZMA_DATA_ERROR;
543 // If we get here and stop is true, something went
544 // wrong and we print an error. Otherwise it's just
545 // a warning and coding can continue.
547 message_error("%s: %s", pair->src_name,
550 message_warning("%s: %s", pair->src_name,
553 // When compressing, all possible errors set
555 assert(opt_mode != MODE_COMPRESS);
558 if (ret == LZMA_MEMLIMIT_ERROR) {
559 // Display how much memory it would have
561 message_mem_needed(V_ERROR,
562 lzma_memusage(&strm));
569 // Show progress information under certain conditions.
570 message_progress_update();
577 /// Copy from input file to output file without processing the data in any
578 /// way. This is used only when trying to decompress unrecognized files
579 /// with --decompress --stdout --force, so the output is always stdout.
581 coder_passthru(file_pair *pair)
583 while (strm.avail_in != 0) {
587 if (io_write(pair, &in_buf, strm.avail_in))
590 strm.total_in += strm.avail_in;
591 strm.total_out = strm.total_in;
592 message_progress_update();
594 strm.avail_in = io_read(pair, &in_buf, IO_BUFFER_SIZE);
595 if (strm.avail_in == SIZE_MAX)
604 coder_run(const char *filename)
606 // Set and possibly print the filename for the progress message.
607 message_filename(filename);
609 // Try to open the input file.
610 file_pair *pair = io_open_src(filename);
614 // Assume that something goes wrong.
615 bool success = false;
617 // Read the first chunk of input data. This is needed to detect
618 // the input file type (for now, only for decompression).
619 strm.next_in = in_buf.u8;
620 strm.avail_in = io_read(pair, &in_buf, IO_BUFFER_SIZE);
622 if (strm.avail_in != SIZE_MAX) {
623 // Initialize the coder. This will detect the file format
624 // and, in decompression or testing mode, check the memory
625 // usage of the first Block too. This way we don't try to
626 // open the destination file if we see that coding wouldn't
627 // work at all anyway. This also avoids deleting the old
628 // "target" file if --force was used.
629 const enum coder_init_ret init_ret = coder_init(pair);
631 if (init_ret != CODER_INIT_ERROR && !user_abort) {
632 // Don't open the destination file when --test
634 if (opt_mode == MODE_TEST || !io_open_dest(pair)) {
635 // Initialize the progress indicator.
636 const uint64_t in_size
637 = pair->src_st.st_size <= 0
638 ? 0 : pair->src_st.st_size;
639 message_progress_start(&strm, in_size);
641 // Do the actual coding or passthru.
642 if (init_ret == CODER_INIT_NORMAL)
643 success = coder_normal(pair);
645 success = coder_passthru(pair);
647 message_progress_end(success);
652 // Close the file pair. It needs to know if coding was successful to
653 // know if the source or target file should be unlinked.
654 io_close(pair, success);