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;
26 enum format_type opt_format = FORMAT_AUTO;
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 /// True if we should auto-adjust the compression settings to use less memory
46 /// if memory usage limit is too low for the original settings.
47 static bool auto_adjust = true;
49 /// Indicate if no preset has been explicitly given. In that case, if we need
50 /// to auto-adjust for lower memory usage, we won't print a warning.
51 static bool preset_default = true;
53 /// If a preset is used (no custom filter chain) and preset_extreme is true,
54 /// a significantly slower compression is used to achieve slightly better
55 /// compression ratio.
56 static bool preset_extreme = false;
58 /// Integrity check type
59 #ifdef HAVE_CHECK_CRC64
60 static lzma_check check = LZMA_CHECK_CRC64;
62 static lzma_check check = LZMA_CHECK_CRC32;
67 coder_set_check(lzma_check new_check)
75 coder_set_preset(size_t new_preset)
77 preset_number = new_preset;
78 preset_default = false;
84 coder_set_extreme(void)
86 preset_extreme = true;
92 coder_add_filter(lzma_vli id, void *options)
94 if (filters_count == LZMA_FILTERS_MAX)
95 message_fatal(_("Maximum number of filters is four"));
97 filters[filters_count].id = id;
98 filters[filters_count].options = options;
105 static void lzma_attribute((noreturn))
106 memlimit_too_small(uint64_t memory_usage)
108 message(V_ERROR, _("Memory usage limit is too low for the given "
110 message_mem_needed(V_ERROR, memory_usage);
111 tuklib_exit(E_ERROR, E_ERROR, false);
116 coder_set_compression_settings(void)
118 // Options for LZMA1 or LZMA2 in case we are using a preset.
119 static lzma_options_lzma opt_lzma;
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 "
131 message(V_WARNING, _("The exact options of the "
132 "presets may vary between software "
136 // Get the preset for LZMA1 or LZMA2.
138 preset_number |= LZMA_PRESET_EXTREME;
140 if (lzma_lzma_preset(&opt_lzma, preset_number))
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;
149 preset_default = false;
152 // Terminate the filter options array.
153 filters[filters_count].id = LZMA_VLI_UNKNOWN;
155 // If we are using the .lzma format, allow exactly one filter
156 // which has to be LZMA1.
157 if (opt_format == FORMAT_LZMA && (filters_count != 1
158 || filters[0].id != LZMA_FILTER_LZMA1))
159 message_fatal(_("The .lzma format supports only "
160 "the LZMA1 filter"));
162 // If we are using the .xz format, make sure that there is no LZMA1
163 // filter to prevent LZMA_PROG_ERROR.
164 if (opt_format == FORMAT_XZ)
165 for (size_t i = 0; i < filters_count; ++i)
166 if (filters[i].id == LZMA_FILTER_LZMA1)
167 message_fatal(_("LZMA1 cannot be used "
168 "with the .xz format"));
170 // Print the selected filter chain.
171 message_filters(V_DEBUG, filters);
173 // If using --format=raw, we can be decoding. The memusage function
174 // also validates the filter chain and the options used for the
176 const uint64_t memory_limit = hardware_memlimit_get();
177 uint64_t memory_usage;
178 if (opt_mode == MODE_COMPRESS)
179 memory_usage = lzma_raw_encoder_memusage(filters);
181 memory_usage = lzma_raw_decoder_memusage(filters);
183 if (memory_usage == UINT64_MAX)
184 message_fatal(_("Unsupported filter chain or filter options"));
186 // Print memory usage info before possible dictionary
187 // size auto-adjusting.
188 message_mem_needed(V_DEBUG, memory_usage);
190 if (memory_usage > memory_limit) {
191 // If --no-auto-adjust was used or we didn't find LZMA1 or
192 // LZMA2 as the last filter, give an error immediatelly.
193 // --format=raw implies --no-auto-adjust.
194 if (!auto_adjust || opt_format == FORMAT_RAW)
195 memlimit_too_small(memory_usage);
197 assert(opt_mode == MODE_COMPRESS);
199 // Look for the last filter if it is LZMA2 or LZMA1, so
200 // we can make it use less RAM. With other filters we don't
203 while (filters[i].id != LZMA_FILTER_LZMA2
204 && filters[i].id != LZMA_FILTER_LZMA1) {
205 if (filters[i].id == LZMA_VLI_UNKNOWN)
206 memlimit_too_small(memory_usage);
211 // Decrease the dictionary size until we meet the memory
212 // usage limit. First round down to full mebibytes.
213 lzma_options_lzma *opt = filters[i].options;
214 const uint32_t orig_dict_size = opt->dict_size;
215 opt->dict_size &= ~((UINT32_C(1) << 20) - 1);
217 // If it is below 1 MiB, auto-adjusting failed. We
218 // could be more sophisticated and scale it down even
219 // more, but let's see if many complain about this
222 // FIXME: Displays the scaled memory usage instead
224 if (opt->dict_size < (UINT32_C(1) << 20))
225 memlimit_too_small(memory_usage);
227 memory_usage = lzma_raw_encoder_memusage(filters);
228 if (memory_usage == UINT64_MAX)
231 // Accept it if it is low enough.
232 if (memory_usage <= memory_limit)
235 // Otherwise 1 MiB down and try again. I hope this
236 // isn't too slow method for cases where the original
237 // dict_size is very big.
238 opt->dict_size -= UINT32_C(1) << 20;
241 // Tell the user that we decreased the dictionary size.
242 // However, omit the message if no preset or custom chain
243 // was given. FIXME: Always warn?
245 message(V_WARNING, _("Adjusted LZMA%c dictionary size "
246 "from %s MiB to %s MiB to not exceed "
247 "the memory usage limit of %s MiB"),
248 filters[i].id == LZMA_FILTER_LZMA2
250 uint64_to_str(orig_dict_size >> 20, 0),
251 uint64_to_str(opt->dict_size >> 20, 1),
252 uint64_to_str(round_up_to_mib(
257 // Limit the number of worker threads so that memory usage
258 // limit isn't exceeded.
259 assert(memory_usage > 0);
260 size_t thread_limit = memory_limit / memory_usage;
261 if (thread_limit == 0)
264 if (opt_threads > thread_limit)
265 opt_threads = thread_limit;
272 /// Return true if the data in in_buf seems to be in the .xz format.
276 return strm.avail_in >= 6 && memcmp(in_buf.u8, "\3757zXZ", 6) == 0;
280 /// Return true if the data in in_buf seems to be in the .lzma format.
284 // The .lzma header is 13 bytes.
285 if (strm.avail_in < 13)
288 // Decode the LZMA1 properties.
289 lzma_filter filter = { .id = LZMA_FILTER_LZMA1 };
290 if (lzma_properties_decode(&filter, NULL, in_buf.u8, 5) != LZMA_OK)
293 // A hack to ditch tons of false positives: We allow only dictionary
294 // sizes that are 2^n or 2^n + 2^(n-1) or UINT32_MAX. LZMA_Alone
295 // created only files with 2^n, but accepts any dictionary size.
296 // If someone complains, this will be reconsidered.
297 lzma_options_lzma *opt = filter.options;
298 const uint32_t dict_size = opt->dict_size;
301 if (dict_size != UINT32_MAX) {
302 uint32_t d = dict_size - 1;
309 if (d != dict_size || dict_size == 0)
313 // Another hack to ditch false positives: Assume that if the
314 // uncompressed size is known, it must be less than 256 GiB.
315 // Again, if someone complains, this will be reconsidered.
316 uint64_t uncompressed_size = 0;
317 for (size_t i = 0; i < 8; ++i)
318 uncompressed_size |= (uint64_t)(in_buf.u8[5 + i]) << (i * 8);
320 if (uncompressed_size != UINT64_MAX
321 && uncompressed_size > (UINT64_C(1) << 38))
328 /// Detect the input file type (for now, this done only when decompressing),
329 /// and initialize an appropriate coder. Return value indicates if a normal
330 /// liblzma-based coder was initialized (CODER_INIT_NORMAL), if passthru
331 /// mode should be used (CODER_INIT_PASSTHRU), or if an error occurred
332 /// (CODER_INIT_ERROR).
333 static enum coder_init_ret
334 coder_init(file_pair *pair)
336 lzma_ret ret = LZMA_PROG_ERROR;
338 if (opt_mode == MODE_COMPRESS) {
339 switch (opt_format) {
341 // args.c ensures this.
346 ret = lzma_stream_encoder(&strm, filters, check);
350 ret = lzma_alone_encoder(&strm, filters[0].options);
354 ret = lzma_raw_encoder(&strm, filters);
358 const uint32_t flags = LZMA_TELL_UNSUPPORTED_CHECK
361 // We abuse FORMAT_AUTO to indicate unknown file format,
362 // for which we may consider passthru mode.
363 enum format_type init_format = FORMAT_AUTO;
365 switch (opt_format) {
368 init_format = FORMAT_XZ;
369 else if (is_format_lzma())
370 init_format = FORMAT_LZMA;
375 init_format = FORMAT_XZ;
379 if (is_format_lzma())
380 init_format = FORMAT_LZMA;
384 init_format = FORMAT_RAW;
388 switch (init_format) {
390 // Uknown file format. If --decompress --stdout
391 // --force have been given, then we copy the input
392 // as is to stdout. Checking for MODE_DECOMPRESS
393 // is needed, because we don't want to do use
394 // passthru mode with --test.
395 if (opt_mode == MODE_DECOMPRESS
396 && opt_stdout && opt_force)
397 return CODER_INIT_PASSTHRU;
399 ret = LZMA_FORMAT_ERROR;
403 ret = lzma_stream_decoder(&strm,
404 hardware_memlimit_get(), flags);
408 ret = lzma_alone_decoder(&strm,
409 hardware_memlimit_get());
413 // Memory usage has already been checked in
414 // coder_set_compression_settings().
415 ret = lzma_raw_decoder(&strm, filters);
419 // Try to decode the headers. This will catch too low
420 // memory usage limit in case it happens in the first
421 // Block of the first Stream, which is where it very
422 // probably will happen if it is going to happen.
423 if (ret == LZMA_OK && init_format != FORMAT_RAW) {
424 strm.next_out = NULL;
426 ret = lzma_code(&strm, LZMA_RUN);
430 if (ret != LZMA_OK) {
431 message_error("%s: %s", pair->src_name, message_strm(ret));
432 if (ret == LZMA_MEMLIMIT_ERROR)
433 message_mem_needed(V_ERROR, lzma_memusage(&strm));
435 return CODER_INIT_ERROR;
438 return CODER_INIT_NORMAL;
442 /// Compress or decompress using liblzma.
444 coder_normal(file_pair *pair)
446 // Encoder needs to know when we have given all the input to it.
447 // The decoders need to know it too when we are using
448 // LZMA_CONCATENATED. We need to check for src_eof here, because
449 // the first input chunk has been already read, and that may
450 // have been the only chunk we will read.
451 lzma_action action = pair->src_eof ? LZMA_FINISH : LZMA_RUN;
455 // Assume that something goes wrong.
456 bool success = false;
458 strm.next_out = out_buf.u8;
459 strm.avail_out = IO_BUFFER_SIZE;
461 while (!user_abort) {
462 // Fill the input buffer if it is empty and we haven't reached
464 if (strm.avail_in == 0 && !pair->src_eof) {
465 strm.next_in = in_buf.u8;
466 strm.avail_in = io_read(
467 pair, &in_buf, IO_BUFFER_SIZE);
469 if (strm.avail_in == SIZE_MAX)
473 action = LZMA_FINISH;
476 // Let liblzma do the actual work.
477 ret = lzma_code(&strm, action);
479 // Write out if the output buffer became full.
480 if (strm.avail_out == 0) {
481 if (opt_mode != MODE_TEST && io_write(pair, &out_buf,
482 IO_BUFFER_SIZE - strm.avail_out))
485 strm.next_out = out_buf.u8;
486 strm.avail_out = IO_BUFFER_SIZE;
489 if (ret != LZMA_OK) {
490 // Determine if the return value indicates that we
491 // won't continue coding.
492 const bool stop = ret != LZMA_NO_CHECK
493 && ret != LZMA_UNSUPPORTED_CHECK;
496 // Write the remaining bytes even if something
497 // went wrong, because that way the user gets
498 // as much data as possible, which can be good
499 // when trying to get at least some useful
500 // data out of damaged files.
501 if (opt_mode != MODE_TEST && io_write(pair,
502 &out_buf, IO_BUFFER_SIZE
507 if (ret == LZMA_STREAM_END) {
508 // Check that there is no trailing garbage.
509 // This is needed for LZMA_Alone and raw
511 if (strm.avail_in == 0 && !pair->src_eof) {
512 // Try reading one more byte.
513 // Hopefully we don't get any more
514 // input, and thus pair->src_eof
516 strm.avail_in = io_read(
518 if (strm.avail_in == SIZE_MAX)
521 assert(strm.avail_in == 0
522 || strm.avail_in == 1);
525 if (strm.avail_in == 0) {
526 assert(pair->src_eof);
531 // We hadn't reached the end of the file.
532 ret = LZMA_DATA_ERROR;
536 // If we get here and stop is true, something went
537 // wrong and we print an error. Otherwise it's just
538 // a warning and coding can continue.
540 message_error("%s: %s", pair->src_name,
543 message_warning("%s: %s", pair->src_name,
546 // When compressing, all possible errors set
548 assert(opt_mode != MODE_COMPRESS);
551 if (ret == LZMA_MEMLIMIT_ERROR) {
552 // Display how much memory it would have
554 message_mem_needed(V_ERROR,
555 lzma_memusage(&strm));
562 // Show progress information under certain conditions.
563 message_progress_update();
570 /// Copy from input file to output file without processing the data in any
571 /// way. This is used only when trying to decompress unrecognized files
572 /// with --decompress --stdout --force, so the output is always stdout.
574 coder_passthru(file_pair *pair)
576 while (strm.avail_in != 0) {
580 if (io_write(pair, &in_buf, strm.avail_in))
583 strm.total_in += strm.avail_in;
584 strm.total_out = strm.total_in;
585 message_progress_update();
587 strm.avail_in = io_read(pair, &in_buf, IO_BUFFER_SIZE);
588 if (strm.avail_in == SIZE_MAX)
597 coder_run(const char *filename)
599 // Set and possibly print the filename for the progress message.
600 message_filename(filename);
602 // Try to open the input file.
603 file_pair *pair = io_open_src(filename);
607 // Assume that something goes wrong.
608 bool success = false;
610 // Read the first chunk of input data. This is needed to detect
611 // the input file type (for now, only for decompression).
612 strm.next_in = in_buf.u8;
613 strm.avail_in = io_read(pair, &in_buf, IO_BUFFER_SIZE);
615 if (strm.avail_in != SIZE_MAX) {
616 // Initialize the coder. This will detect the file format
617 // and, in decompression or testing mode, check the memory
618 // usage of the first Block too. This way we don't try to
619 // open the destination file if we see that coding wouldn't
620 // work at all anyway. This also avoids deleting the old
621 // "target" file if --force was used.
622 const enum coder_init_ret init_ret = coder_init(pair);
624 if (init_ret != CODER_INIT_ERROR && !user_abort) {
625 // Don't open the destination file when --test
627 if (opt_mode == MODE_TEST || !io_open_dest(pair)) {
628 // Initialize the progress indicator.
629 const uint64_t in_size
630 = pair->src_st.st_size <= 0
631 ? 0 : pair->src_st.st_size;
632 message_progress_start(&strm, in_size);
634 // Do the actual coding or passthru.
635 if (init_ret == CODER_INIT_NORMAL)
636 success = coder_normal(pair);
638 success = coder_passthru(pair);
640 message_progress_end(success);
645 // Close the file pair. It needs to know if coding was successful to
646 // know if the source or target file should be unlinked.
647 io_close(pair, success);