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 uint32_t filters_count = 0;
42 /// Number of the preset (0-9)
43 static uint32_t preset_number = LZMA_PRESET_DEFAULT;
45 /// Integrity check type
46 static lzma_check check;
48 /// This becomes false if the --check=CHECK option is used.
49 static bool check_default = true;
53 coder_set_check(lzma_check new_check)
56 check_default = false;
62 forget_filter_chain(void)
64 // Setting a preset makes us forget a possibly defined custom
66 while (filters_count > 0) {
68 free(filters[filters_count].options);
69 filters[filters_count].options = NULL;
77 coder_set_preset(uint32_t new_preset)
79 preset_number &= ~LZMA_PRESET_LEVEL_MASK;
80 preset_number |= new_preset;
81 forget_filter_chain();
87 coder_set_extreme(void)
89 preset_number |= LZMA_PRESET_EXTREME;
90 forget_filter_chain();
96 coder_add_filter(lzma_vli id, void *options)
98 if (filters_count == LZMA_FILTERS_MAX)
99 message_fatal(_("Maximum number of filters is four"));
101 filters[filters_count].id = id;
102 filters[filters_count].options = options;
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;
115 static void lzma_attribute((__noreturn__))
116 memlimit_too_small(uint64_t memory_usage)
118 message(V_ERROR, _("Memory usage limit is too low for the given "
120 message_mem_needed(V_ERROR, memory_usage);
121 tuklib_exit(E_ERROR, E_ERROR, false);
126 coder_set_compression_settings(void)
128 // Options for LZMA1 or LZMA2 in case we are using a preset.
129 static lzma_options_lzma opt_lzma;
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 "
141 message(V_WARNING, _("The exact options of the "
142 "presets may vary between software "
146 // Get the preset for LZMA1 or LZMA2.
147 if (lzma_lzma_preset(&opt_lzma, preset_number))
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;
157 // Terminate the filter options array.
158 filters[filters_count].id = LZMA_VLI_UNKNOWN;
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"));
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"));
175 // Print the selected filter chain.
176 message_filters_show(V_DEBUG, filters);
178 // If using --format=raw, we can be decoding. The memusage function
179 // also validates the filter chain and the options used for the
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);
186 memory_usage = lzma_raw_decoder_memusage(filters);
188 if (memory_usage == UINT64_MAX)
189 message_fatal(_("Unsupported filter chain or filter options"));
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));
202 if (memory_usage > memory_limit) {
203 // If --no-auto-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-auto-adjust.
206 if (!opt_auto_adjust || opt_format == FORMAT_RAW)
207 memlimit_too_small(memory_usage);
209 assert(opt_mode == MODE_COMPRESS);
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
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);
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);
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
234 // FIXME: Displays the scaled memory usage instead
236 if (opt->dict_size < (UINT32_C(1) << 20))
237 memlimit_too_small(memory_usage);
239 memory_usage = lzma_raw_encoder_memusage(filters);
240 if (memory_usage == UINT64_MAX)
243 // Accept it if it is low enough.
244 if (memory_usage <= memory_limit)
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;
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
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(
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)
273 if (opt_threads > thread_limit)
274 opt_threads = thread_limit;
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;
290 /// Return true if the data in in_buf seems to be in the .xz format.
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;
301 /// Return true if the data in in_buf seems to be in the .lzma format.
305 // The .lzma header is 13 bytes.
306 if (strm.avail_in < 13)
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)
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;
322 if (dict_size != UINT32_MAX) {
323 uint32_t d = dict_size - 1;
330 if (d != dict_size || dict_size == 0)
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);
341 if (uncompressed_size != UINT64_MAX
342 && uncompressed_size > (UINT64_C(1) << 38))
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).
354 static enum coder_init_ret
355 coder_init(file_pair *pair)
357 lzma_ret ret = LZMA_PROG_ERROR;
359 if (opt_mode == MODE_COMPRESS) {
360 switch (opt_format) {
362 // args.c ensures this.
367 ret = lzma_stream_encoder(&strm, filters, check);
371 ret = lzma_alone_encoder(&strm, filters[0].options);
375 ret = lzma_raw_encoder(&strm, filters);
379 const uint32_t flags = LZMA_TELL_UNSUPPORTED_CHECK
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;
386 switch (opt_format) {
389 init_format = FORMAT_XZ;
390 else if (is_format_lzma())
391 init_format = FORMAT_LZMA;
396 init_format = FORMAT_XZ;
400 if (is_format_lzma())
401 init_format = FORMAT_LZMA;
405 init_format = FORMAT_RAW;
409 switch (init_format) {
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;
420 ret = LZMA_FORMAT_ERROR;
424 ret = lzma_stream_decoder(&strm,
425 hardware_memlimit_get(
426 MODE_DECOMPRESS), flags);
430 ret = lzma_alone_decoder(&strm,
431 hardware_memlimit_get(
436 // Memory usage has already been checked in
437 // coder_set_compression_settings().
438 ret = lzma_raw_decoder(&strm, filters);
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;
449 ret = lzma_code(&strm, LZMA_RUN);
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));
458 return CODER_INIT_ERROR;
461 return CODER_INIT_NORMAL;
465 /// Compress or decompress using liblzma.
467 coder_normal(file_pair *pair)
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;
478 // Assume that something goes wrong.
479 bool success = false;
481 strm.next_out = out_buf.u8;
482 strm.avail_out = IO_BUFFER_SIZE;
484 while (!user_abort) {
485 // Fill the input buffer if it is empty and we haven't reached
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);
492 if (strm.avail_in == SIZE_MAX)
496 action = LZMA_FINISH;
499 // Let liblzma do the actual work.
500 ret = lzma_code(&strm, action);
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))
508 strm.next_out = out_buf.u8;
509 strm.avail_out = IO_BUFFER_SIZE;
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;
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
530 if (ret == LZMA_STREAM_END) {
531 // Check that there is no trailing garbage.
532 // This is needed for LZMA_Alone and raw
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
539 strm.avail_in = io_read(
541 if (strm.avail_in == SIZE_MAX)
544 assert(strm.avail_in == 0
545 || strm.avail_in == 1);
548 if (strm.avail_in == 0) {
549 assert(pair->src_eof);
554 // We hadn't reached the end of the file.
555 ret = LZMA_DATA_ERROR;
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.
563 message_error("%s: %s", pair->src_name,
566 message_warning("%s: %s", pair->src_name,
569 // When compressing, all possible errors set
571 assert(opt_mode != MODE_COMPRESS);
574 if (ret == LZMA_MEMLIMIT_ERROR) {
575 // Display how much memory it would have
577 message_mem_needed(V_ERROR,
578 lzma_memusage(&strm));
585 // Show progress information under certain conditions.
586 message_progress_update();
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.
597 coder_passthru(file_pair *pair)
599 while (strm.avail_in != 0) {
603 if (io_write(pair, &in_buf, strm.avail_in))
606 strm.total_in += strm.avail_in;
607 strm.total_out = strm.total_in;
608 message_progress_update();
610 strm.avail_in = io_read(pair, &in_buf, IO_BUFFER_SIZE);
611 if (strm.avail_in == SIZE_MAX)
620 coder_run(const char *filename)
622 // Set and possibly print the filename for the progress message.
623 message_filename(filename);
625 // Try to open the input file.
626 file_pair *pair = io_open_src(filename);
630 // Assume that something goes wrong.
631 bool success = false;
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);
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);
647 if (init_ret != CODER_INIT_ERROR && !user_abort) {
648 // Don't open the destination file when --test
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);
657 // Do the actual coding or passthru.
658 if (init_ret == CODER_INIT_NORMAL)
659 success = coder_normal(pair);
661 success = coder_passthru(pair);
663 message_progress_end(success);
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);