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 /// 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;
50 /// Integrity check type
51 static lzma_check check;
53 /// This becomes false if the --check=CHECK option is used.
54 static bool check_default = true;
58 coder_set_check(lzma_check new_check)
61 check_default = false;
67 coder_set_preset(size_t new_preset)
69 preset_number = new_preset;
75 coder_set_extreme(void)
77 preset_extreme = true;
83 coder_add_filter(lzma_vli id, void *options)
85 if (filters_count == LZMA_FILTERS_MAX)
86 message_fatal(_("Maximum number of filters is four"));
88 filters[filters_count].id = id;
89 filters[filters_count].options = options;
96 static void lzma_attribute((noreturn))
97 memlimit_too_small(uint64_t memory_usage)
99 message(V_ERROR, _("Memory usage limit is too low for the given "
101 message_mem_needed(V_ERROR, memory_usage);
102 tuklib_exit(E_ERROR, E_ERROR, false);
107 coder_set_compression_settings(void)
109 // Options for LZMA1 or LZMA2 in case we are using a preset.
110 static lzma_options_lzma opt_lzma;
112 if (filters_count == 0) {
113 // We are using a preset. This is not a good idea in raw mode
114 // except when playing around with things. Different versions
115 // of this software may use different options in presets, and
116 // thus make uncompressing the raw data difficult.
117 if (opt_format == FORMAT_RAW) {
118 // The message is shown only if warnings are allowed
119 // but the exit status isn't changed.
120 message(V_WARNING, _("Using a preset in raw mode "
122 message(V_WARNING, _("The exact options of the "
123 "presets may vary between software "
127 // Get the preset for LZMA1 or LZMA2.
129 preset_number |= LZMA_PRESET_EXTREME;
131 if (lzma_lzma_preset(&opt_lzma, preset_number))
134 // Use LZMA2 except with --format=lzma we use LZMA1.
135 filters[0].id = opt_format == FORMAT_LZMA
136 ? LZMA_FILTER_LZMA1 : LZMA_FILTER_LZMA2;
137 filters[0].options = &opt_lzma;
141 // Terminate the filter options array.
142 filters[filters_count].id = LZMA_VLI_UNKNOWN;
144 // If we are using the .lzma format, allow exactly one filter
145 // which has to be LZMA1.
146 if (opt_format == FORMAT_LZMA && (filters_count != 1
147 || filters[0].id != LZMA_FILTER_LZMA1))
148 message_fatal(_("The .lzma format supports only "
149 "the LZMA1 filter"));
151 // If we are using the .xz format, make sure that there is no LZMA1
152 // filter to prevent LZMA_PROG_ERROR.
153 if (opt_format == FORMAT_XZ)
154 for (size_t i = 0; i < filters_count; ++i)
155 if (filters[i].id == LZMA_FILTER_LZMA1)
156 message_fatal(_("LZMA1 cannot be used "
157 "with the .xz format"));
159 // Print the selected filter chain.
160 message_filters_show(V_DEBUG, filters);
162 // If using --format=raw, we can be decoding. The memusage function
163 // also validates the filter chain and the options used for the
165 const uint64_t memory_limit = hardware_memlimit_get(opt_mode);
166 uint64_t memory_usage;
167 if (opt_mode == MODE_COMPRESS)
168 memory_usage = lzma_raw_encoder_memusage(filters);
170 memory_usage = lzma_raw_decoder_memusage(filters);
172 if (memory_usage == UINT64_MAX)
173 message_fatal(_("Unsupported filter chain or filter options"));
175 // Print memory usage info before possible dictionary
176 // size auto-adjusting.
177 message_mem_needed(V_DEBUG, memory_usage);
179 if (memory_usage > memory_limit) {
180 // If --no-auto-adjust was used or we didn't find LZMA1 or
181 // LZMA2 as the last filter, give an error immediately.
182 // --format=raw implies --no-auto-adjust.
183 if (!opt_auto_adjust || opt_format == FORMAT_RAW)
184 memlimit_too_small(memory_usage);
186 assert(opt_mode == MODE_COMPRESS);
188 // Look for the last filter if it is LZMA2 or LZMA1, so
189 // we can make it use less RAM. With other filters we don't
192 while (filters[i].id != LZMA_FILTER_LZMA2
193 && filters[i].id != LZMA_FILTER_LZMA1) {
194 if (filters[i].id == LZMA_VLI_UNKNOWN)
195 memlimit_too_small(memory_usage);
200 // Decrease the dictionary size until we meet the memory
201 // usage limit. First round down to full mebibytes.
202 lzma_options_lzma *opt = filters[i].options;
203 const uint32_t orig_dict_size = opt->dict_size;
204 opt->dict_size &= ~((UINT32_C(1) << 20) - 1);
206 // If it is below 1 MiB, auto-adjusting failed. We
207 // could be more sophisticated and scale it down even
208 // more, but let's see if many complain about this
211 // FIXME: Displays the scaled memory usage instead
213 if (opt->dict_size < (UINT32_C(1) << 20))
214 memlimit_too_small(memory_usage);
216 memory_usage = lzma_raw_encoder_memusage(filters);
217 if (memory_usage == UINT64_MAX)
220 // Accept it if it is low enough.
221 if (memory_usage <= memory_limit)
224 // Otherwise 1 MiB down and try again. I hope this
225 // isn't too slow method for cases where the original
226 // dict_size is very big.
227 opt->dict_size -= UINT32_C(1) << 20;
230 // Tell the user that we decreased the dictionary size.
231 message(V_WARNING, _("Adjusted LZMA%c dictionary size "
232 "from %s MiB to %s MiB to not exceed "
233 "the memory usage limit of %s MiB"),
234 filters[i].id == LZMA_FILTER_LZMA2
236 uint64_to_str(orig_dict_size >> 20, 0),
237 uint64_to_str(opt->dict_size >> 20, 1),
238 uint64_to_str(round_up_to_mib(
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)
250 if (opt_threads > thread_limit)
251 opt_threads = thread_limit;
255 // The default check type is CRC64, but fallback to CRC32
256 // if CRC64 isn't supported by the copy of liblzma we are
257 // using. CRC32 is always supported.
258 check = LZMA_CHECK_CRC64;
259 if (!lzma_check_is_supported(check))
260 check = LZMA_CHECK_CRC32;
267 /// Return true if the data in in_buf seems to be in the .xz format.
271 return strm.avail_in >= 6 && memcmp(in_buf.u8, "\3757zXZ", 6) == 0;
275 /// Return true if the data in in_buf seems to be in the .lzma format.
279 // The .lzma header is 13 bytes.
280 if (strm.avail_in < 13)
283 // Decode the LZMA1 properties.
284 lzma_filter filter = { .id = LZMA_FILTER_LZMA1 };
285 if (lzma_properties_decode(&filter, NULL, in_buf.u8, 5) != LZMA_OK)
288 // A hack to ditch tons of false positives: We allow only dictionary
289 // sizes that are 2^n or 2^n + 2^(n-1) or UINT32_MAX. LZMA_Alone
290 // created only files with 2^n, but accepts any dictionary size.
291 // If someone complains, this will be reconsidered.
292 lzma_options_lzma *opt = filter.options;
293 const uint32_t dict_size = opt->dict_size;
296 if (dict_size != UINT32_MAX) {
297 uint32_t d = dict_size - 1;
304 if (d != dict_size || dict_size == 0)
308 // Another hack to ditch false positives: Assume that if the
309 // uncompressed size is known, it must be less than 256 GiB.
310 // Again, if someone complains, this will be reconsidered.
311 uint64_t uncompressed_size = 0;
312 for (size_t i = 0; i < 8; ++i)
313 uncompressed_size |= (uint64_t)(in_buf.u8[5 + i]) << (i * 8);
315 if (uncompressed_size != UINT64_MAX
316 && uncompressed_size > (UINT64_C(1) << 38))
323 /// Detect the input file type (for now, this done only when decompressing),
324 /// and initialize an appropriate coder. Return value indicates if a normal
325 /// liblzma-based coder was initialized (CODER_INIT_NORMAL), if passthru
326 /// mode should be used (CODER_INIT_PASSTHRU), or if an error occurred
327 /// (CODER_INIT_ERROR).
328 static enum coder_init_ret
329 coder_init(file_pair *pair)
331 lzma_ret ret = LZMA_PROG_ERROR;
333 if (opt_mode == MODE_COMPRESS) {
334 switch (opt_format) {
336 // args.c ensures this.
341 ret = lzma_stream_encoder(&strm, filters, check);
345 ret = lzma_alone_encoder(&strm, filters[0].options);
349 ret = lzma_raw_encoder(&strm, filters);
353 const uint32_t flags = LZMA_TELL_UNSUPPORTED_CHECK
356 // We abuse FORMAT_AUTO to indicate unknown file format,
357 // for which we may consider passthru mode.
358 enum format_type init_format = FORMAT_AUTO;
360 switch (opt_format) {
363 init_format = FORMAT_XZ;
364 else if (is_format_lzma())
365 init_format = FORMAT_LZMA;
370 init_format = FORMAT_XZ;
374 if (is_format_lzma())
375 init_format = FORMAT_LZMA;
379 init_format = FORMAT_RAW;
383 switch (init_format) {
385 // Uknown file format. If --decompress --stdout
386 // --force have been given, then we copy the input
387 // as is to stdout. Checking for MODE_DECOMPRESS
388 // is needed, because we don't want to do use
389 // passthru mode with --test.
390 if (opt_mode == MODE_DECOMPRESS
391 && opt_stdout && opt_force)
392 return CODER_INIT_PASSTHRU;
394 ret = LZMA_FORMAT_ERROR;
398 ret = lzma_stream_decoder(&strm,
399 hardware_memlimit_get(
400 MODE_DECOMPRESS), flags);
404 ret = lzma_alone_decoder(&strm,
405 hardware_memlimit_get(
410 // Memory usage has already been checked in
411 // coder_set_compression_settings().
412 ret = lzma_raw_decoder(&strm, filters);
416 // Try to decode the headers. This will catch too low
417 // memory usage limit in case it happens in the first
418 // Block of the first Stream, which is where it very
419 // probably will happen if it is going to happen.
420 if (ret == LZMA_OK && init_format != FORMAT_RAW) {
421 strm.next_out = NULL;
423 ret = lzma_code(&strm, LZMA_RUN);
427 if (ret != LZMA_OK) {
428 message_error("%s: %s", pair->src_name, message_strm(ret));
429 if (ret == LZMA_MEMLIMIT_ERROR)
430 message_mem_needed(V_ERROR, lzma_memusage(&strm));
432 return CODER_INIT_ERROR;
435 return CODER_INIT_NORMAL;
439 /// Compress or decompress using liblzma.
441 coder_normal(file_pair *pair)
443 // Encoder needs to know when we have given all the input to it.
444 // The decoders need to know it too when we are using
445 // LZMA_CONCATENATED. We need to check for src_eof here, because
446 // the first input chunk has been already read, and that may
447 // have been the only chunk we will read.
448 lzma_action action = pair->src_eof ? LZMA_FINISH : LZMA_RUN;
452 // Assume that something goes wrong.
453 bool success = false;
455 strm.next_out = out_buf.u8;
456 strm.avail_out = IO_BUFFER_SIZE;
458 while (!user_abort) {
459 // Fill the input buffer if it is empty and we haven't reached
461 if (strm.avail_in == 0 && !pair->src_eof) {
462 strm.next_in = in_buf.u8;
463 strm.avail_in = io_read(
464 pair, &in_buf, IO_BUFFER_SIZE);
466 if (strm.avail_in == SIZE_MAX)
470 action = LZMA_FINISH;
473 // Let liblzma do the actual work.
474 ret = lzma_code(&strm, action);
476 // Write out if the output buffer became full.
477 if (strm.avail_out == 0) {
478 if (opt_mode != MODE_TEST && io_write(pair, &out_buf,
479 IO_BUFFER_SIZE - strm.avail_out))
482 strm.next_out = out_buf.u8;
483 strm.avail_out = IO_BUFFER_SIZE;
486 if (ret != LZMA_OK) {
487 // Determine if the return value indicates that we
488 // won't continue coding.
489 const bool stop = ret != LZMA_NO_CHECK
490 && ret != LZMA_UNSUPPORTED_CHECK;
493 // Write the remaining bytes even if something
494 // went wrong, because that way the user gets
495 // as much data as possible, which can be good
496 // when trying to get at least some useful
497 // data out of damaged files.
498 if (opt_mode != MODE_TEST && io_write(pair,
499 &out_buf, IO_BUFFER_SIZE
504 if (ret == LZMA_STREAM_END) {
505 // Check that there is no trailing garbage.
506 // This is needed for LZMA_Alone and raw
508 if (strm.avail_in == 0 && !pair->src_eof) {
509 // Try reading one more byte.
510 // Hopefully we don't get any more
511 // input, and thus pair->src_eof
513 strm.avail_in = io_read(
515 if (strm.avail_in == SIZE_MAX)
518 assert(strm.avail_in == 0
519 || strm.avail_in == 1);
522 if (strm.avail_in == 0) {
523 assert(pair->src_eof);
528 // We hadn't reached the end of the file.
529 ret = LZMA_DATA_ERROR;
533 // If we get here and stop is true, something went
534 // wrong and we print an error. Otherwise it's just
535 // a warning and coding can continue.
537 message_error("%s: %s", pair->src_name,
540 message_warning("%s: %s", pair->src_name,
543 // When compressing, all possible errors set
545 assert(opt_mode != MODE_COMPRESS);
548 if (ret == LZMA_MEMLIMIT_ERROR) {
549 // Display how much memory it would have
551 message_mem_needed(V_ERROR,
552 lzma_memusage(&strm));
559 // Show progress information under certain conditions.
560 message_progress_update();
567 /// Copy from input file to output file without processing the data in any
568 /// way. This is used only when trying to decompress unrecognized files
569 /// with --decompress --stdout --force, so the output is always stdout.
571 coder_passthru(file_pair *pair)
573 while (strm.avail_in != 0) {
577 if (io_write(pair, &in_buf, strm.avail_in))
580 strm.total_in += strm.avail_in;
581 strm.total_out = strm.total_in;
582 message_progress_update();
584 strm.avail_in = io_read(pair, &in_buf, IO_BUFFER_SIZE);
585 if (strm.avail_in == SIZE_MAX)
594 coder_run(const char *filename)
596 // Set and possibly print the filename for the progress message.
597 message_filename(filename);
599 // Try to open the input file.
600 file_pair *pair = io_open_src(filename);
604 // Assume that something goes wrong.
605 bool success = false;
607 // Read the first chunk of input data. This is needed to detect
608 // the input file type (for now, only for decompression).
609 strm.next_in = in_buf.u8;
610 strm.avail_in = io_read(pair, &in_buf, IO_BUFFER_SIZE);
612 if (strm.avail_in != SIZE_MAX) {
613 // Initialize the coder. This will detect the file format
614 // and, in decompression or testing mode, check the memory
615 // usage of the first Block too. This way we don't try to
616 // open the destination file if we see that coding wouldn't
617 // work at all anyway. This also avoids deleting the old
618 // "target" file if --force was used.
619 const enum coder_init_ret init_ret = coder_init(pair);
621 if (init_ret != CODER_INIT_ERROR && !user_abort) {
622 // Don't open the destination file when --test
624 if (opt_mode == MODE_TEST || !io_open_dest(pair)) {
625 // Initialize the progress indicator.
626 const uint64_t in_size
627 = pair->src_st.st_size <= 0
628 ? 0 : pair->src_st.st_size;
629 message_progress_start(&strm, in_size);
631 // Do the actual coding or passthru.
632 if (init_ret == CODER_INIT_NORMAL)
633 success = coder_normal(pair);
635 success = coder_passthru(pair);
637 message_progress_end(success);
642 // Close the file pair. It needs to know if coding was successful to
643 // know if the source or target file should be unlinked.
644 io_close(pair, success);