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, uint64_t memory_limit)
108 message_fatal(_("Memory usage limit (%" PRIu64 " MiB) is too small "
109 "for the given filter setup (%" PRIu64 " MiB)"),
110 memory_limit >> 20, memory_usage >> 20);
115 coder_set_compression_settings(void)
117 // Options for LZMA1 or LZMA2 in case we are using a preset.
118 static lzma_options_lzma opt_lzma;
120 if (filters_count == 0) {
121 // We are using a preset. This is not a good idea in raw mode
122 // except when playing around with things. Different versions
123 // of this software may use different options in presets, and
124 // thus make uncompressing the raw data difficult.
125 if (opt_format == FORMAT_RAW) {
126 // The message is shown only if warnings are allowed
127 // but the exit status isn't changed.
128 message(V_WARNING, _("Using a preset in raw mode "
130 message(V_WARNING, _("The exact options of the "
131 "presets may vary between software "
135 // Get the preset for LZMA1 or LZMA2.
137 preset_number |= LZMA_PRESET_EXTREME;
139 if (lzma_lzma_preset(&opt_lzma, preset_number))
142 // Use LZMA2 except with --format=lzma we use LZMA1.
143 filters[0].id = opt_format == FORMAT_LZMA
144 ? LZMA_FILTER_LZMA1 : LZMA_FILTER_LZMA2;
145 filters[0].options = &opt_lzma;
148 preset_default = false;
151 // Terminate the filter options array.
152 filters[filters_count].id = LZMA_VLI_UNKNOWN;
154 // If we are using the .lzma format, allow exactly one filter
155 // which has to be LZMA1.
156 if (opt_format == FORMAT_LZMA && (filters_count != 1
157 || filters[0].id != LZMA_FILTER_LZMA1))
158 message_fatal(_("The .lzma format supports only "
159 "the LZMA1 filter"));
161 // If we are using the .xz format, make sure that there is no LZMA1
162 // filter to prevent LZMA_PROG_ERROR.
163 if (opt_format == FORMAT_XZ)
164 for (size_t i = 0; i < filters_count; ++i)
165 if (filters[i].id == LZMA_FILTER_LZMA1)
166 message_fatal(_("LZMA1 cannot be used "
167 "with the .xz format"));
169 // Print the selected filter chain.
170 message_filters(V_DEBUG, filters);
172 // If using --format=raw, we can be decoding. The memusage function
173 // also validates the filter chain and the options used for the
175 const uint64_t memory_limit = hardware_memlimit_get();
176 uint64_t memory_usage;
177 if (opt_mode == MODE_COMPRESS)
178 memory_usage = lzma_raw_encoder_memusage(filters);
180 memory_usage = lzma_raw_decoder_memusage(filters);
182 if (memory_usage == UINT64_MAX)
183 message_fatal("Unsupported filter chain or filter options");
185 // Print memory usage info.
186 message(V_DEBUG, _("%s MiB (%s B) of memory is required per thread, "
187 "limit is %s MiB (%s B)"),
188 uint64_to_str(memory_usage >> 20, 0),
189 uint64_to_str(memory_usage, 1),
190 uint64_to_str(memory_limit >> 20, 2),
191 uint64_to_str(memory_limit, 3));
193 if (memory_usage > memory_limit) {
194 // If --no-auto-adjust was used or we didn't find LZMA1 or
195 // LZMA2 as the last filter, give an error immediatelly.
196 // --format=raw implies --no-auto-adjust.
197 if (!auto_adjust || opt_format == FORMAT_RAW)
198 memlimit_too_small(memory_usage, memory_limit);
200 assert(opt_mode == MODE_COMPRESS);
202 // Look for the last filter if it is LZMA2 or LZMA1, so
203 // we can make it use less RAM. With other filters we don't
206 while (filters[i].id != LZMA_FILTER_LZMA2
207 && filters[i].id != LZMA_FILTER_LZMA1) {
208 if (filters[i].id == LZMA_VLI_UNKNOWN)
209 memlimit_too_small(memory_usage, memory_limit);
214 // Decrease the dictionary size until we meet the memory
215 // usage limit. First round down to full mebibytes.
216 lzma_options_lzma *opt = filters[i].options;
217 const uint32_t orig_dict_size = opt->dict_size;
218 opt->dict_size &= ~((UINT32_C(1) << 20) - 1);
220 // If it is below 1 MiB, auto-adjusting failed. We
221 // could be more sophisticated and scale it down even
222 // more, but let's see if many complain about this
225 // FIXME: Displays the scaled memory usage instead
227 if (opt->dict_size < (UINT32_C(1) << 20))
228 memlimit_too_small(memory_usage, memory_limit);
230 memory_usage = lzma_raw_encoder_memusage(filters);
231 if (memory_usage == UINT64_MAX)
234 // Accept it if it is low enough.
235 if (memory_usage <= memory_limit)
238 // Otherwise 1 MiB down and try again. I hope this
239 // isn't too slow method for cases where the original
240 // dict_size is very big.
241 opt->dict_size -= UINT32_C(1) << 20;
244 // Tell the user that we decreased the dictionary size.
245 // However, omit the message if no preset or custom chain
246 // was given. FIXME: Always warn?
248 message(V_WARNING, "Adjusted LZMA%c dictionary size "
249 "from %s MiB to %s MiB to not exceed "
250 "the memory usage limit of %s MiB",
251 filters[i].id == LZMA_FILTER_LZMA2
253 uint64_to_str(orig_dict_size >> 20, 0),
254 uint64_to_str(opt->dict_size >> 20, 1),
255 uint64_to_str(memory_limit >> 20, 2));
259 // Limit the number of worker threads so that memory usage
260 // limit isn't exceeded.
261 assert(memory_usage > 0);
262 size_t thread_limit = memory_limit / memory_usage;
263 if (thread_limit == 0)
266 if (opt_threads > thread_limit)
267 opt_threads = thread_limit;
274 /// Return true if the data in in_buf seems to be in the .xz format.
278 return strm.avail_in >= 6 && memcmp(in_buf.u8, "\3757zXZ", 6) == 0;
282 /// Return true if the data in in_buf seems to be in the .lzma format.
286 // The .lzma header is 13 bytes.
287 if (strm.avail_in < 13)
290 // Decode the LZMA1 properties.
291 lzma_filter filter = { .id = LZMA_FILTER_LZMA1 };
292 if (lzma_properties_decode(&filter, NULL, in_buf.u8, 5) != LZMA_OK)
295 // A hack to ditch tons of false positives: We allow only dictionary
296 // sizes that are 2^n or 2^n + 2^(n-1) or UINT32_MAX. LZMA_Alone
297 // created only files with 2^n, but accepts any dictionary size.
298 // If someone complains, this will be reconsidered.
299 lzma_options_lzma *opt = filter.options;
300 const uint32_t dict_size = opt->dict_size;
303 if (dict_size != UINT32_MAX) {
304 uint32_t d = dict_size - 1;
311 if (d != dict_size || dict_size == 0)
315 // Another hack to ditch false positives: Assume that if the
316 // uncompressed size is known, it must be less than 256 GiB.
317 // Again, if someone complains, this will be reconsidered.
318 uint64_t uncompressed_size = 0;
319 for (size_t i = 0; i < 8; ++i)
320 uncompressed_size |= (uint64_t)(in_buf.u8[5 + i]) << (i * 8);
322 if (uncompressed_size != UINT64_MAX
323 && uncompressed_size > (UINT64_C(1) << 38))
330 /// Detect the input file type (for now, this done only when decompressing),
331 /// and initialize an appropriate coder. Return value indicates if a normal
332 /// liblzma-based coder was initialized (CODER_INIT_NORMAL), if passthru
333 /// mode should be used (CODER_INIT_PASSTHRU), or if an error occurred
334 /// (CODER_INIT_ERROR).
335 static enum coder_init_ret
336 coder_init(file_pair *pair)
338 lzma_ret ret = LZMA_PROG_ERROR;
340 if (opt_mode == MODE_COMPRESS) {
341 switch (opt_format) {
343 // args.c ensures this.
348 ret = lzma_stream_encoder(&strm, filters, check);
352 ret = lzma_alone_encoder(&strm, filters[0].options);
356 ret = lzma_raw_encoder(&strm, filters);
360 const uint32_t flags = LZMA_TELL_UNSUPPORTED_CHECK
363 // We abuse FORMAT_AUTO to indicate unknown file format,
364 // for which we may consider passthru mode.
365 enum format_type init_format = FORMAT_AUTO;
367 switch (opt_format) {
370 init_format = FORMAT_XZ;
371 else if (is_format_lzma())
372 init_format = FORMAT_LZMA;
377 init_format = FORMAT_XZ;
381 if (is_format_lzma())
382 init_format = FORMAT_LZMA;
386 init_format = FORMAT_RAW;
390 switch (init_format) {
392 // Uknown file format. If --decompress --stdout
393 // --force have been given, then we copy the input
394 // as is to stdout. Checking for MODE_DECOMPRESS
395 // is needed, because we don't want to do use
396 // passthru mode with --test.
397 if (opt_mode == MODE_DECOMPRESS
398 && opt_stdout && opt_force)
399 return CODER_INIT_PASSTHRU;
401 ret = LZMA_FORMAT_ERROR;
405 ret = lzma_stream_decoder(&strm,
406 hardware_memlimit_get(), flags);
410 ret = lzma_alone_decoder(&strm,
411 hardware_memlimit_get());
415 // Memory usage has already been checked in
416 // coder_set_compression_settings().
417 ret = lzma_raw_decoder(&strm, filters);
422 if (ret != LZMA_OK) {
423 message_error("%s: %s", pair->src_name, message_strm(ret));
424 return CODER_INIT_ERROR;
427 return CODER_INIT_NORMAL;
431 /// Compress or decompress using liblzma.
433 coder_normal(file_pair *pair)
435 // Encoder needs to know when we have given all the input to it.
436 // The decoders need to know it too when we are using
437 // LZMA_CONCATENATED. We need to check for src_eof here, because
438 // the first input chunk has been already read, and that may
439 // have been the only chunk we will read.
440 lzma_action action = pair->src_eof ? LZMA_FINISH : LZMA_RUN;
444 // Assume that something goes wrong.
445 bool success = false;
447 strm.next_out = out_buf.u8;
448 strm.avail_out = IO_BUFFER_SIZE;
450 while (!user_abort) {
451 // Fill the input buffer if it is empty and we haven't reached
453 if (strm.avail_in == 0 && !pair->src_eof) {
454 strm.next_in = in_buf.u8;
455 strm.avail_in = io_read(
456 pair, &in_buf, IO_BUFFER_SIZE);
458 if (strm.avail_in == SIZE_MAX)
462 action = LZMA_FINISH;
465 // Let liblzma do the actual work.
466 ret = lzma_code(&strm, action);
468 // Write out if the output buffer became full.
469 if (strm.avail_out == 0) {
470 if (opt_mode != MODE_TEST && io_write(pair, &out_buf,
471 IO_BUFFER_SIZE - strm.avail_out))
474 strm.next_out = out_buf.u8;
475 strm.avail_out = IO_BUFFER_SIZE;
478 if (ret != LZMA_OK) {
479 // Determine if the return value indicates that we
480 // won't continue coding.
481 const bool stop = ret != LZMA_NO_CHECK
482 && ret != LZMA_UNSUPPORTED_CHECK;
485 // Write the remaining bytes even if something
486 // went wrong, because that way the user gets
487 // as much data as possible, which can be good
488 // when trying to get at least some useful
489 // data out of damaged files.
490 if (opt_mode != MODE_TEST && io_write(pair,
491 &out_buf, IO_BUFFER_SIZE
496 if (ret == LZMA_STREAM_END) {
497 // Check that there is no trailing garbage.
498 // This is needed for LZMA_Alone and raw
500 if (strm.avail_in == 0 && !pair->src_eof) {
501 // Try reading one more byte.
502 // Hopefully we don't get any more
503 // input, and thus pair->src_eof
505 strm.avail_in = io_read(
507 if (strm.avail_in == SIZE_MAX)
510 assert(strm.avail_in == 0
511 || strm.avail_in == 1);
514 if (strm.avail_in == 0) {
515 assert(pair->src_eof);
520 // We hadn't reached the end of the file.
521 ret = LZMA_DATA_ERROR;
525 // If we get here and stop is true, something went
526 // wrong and we print an error. Otherwise it's just
527 // a warning and coding can continue.
529 message_error("%s: %s", pair->src_name,
532 message_warning("%s: %s", pair->src_name,
535 // When compressing, all possible errors set
537 assert(opt_mode != MODE_COMPRESS);
540 if (ret == LZMA_MEMLIMIT_ERROR) {
541 // Figure out how much memory it would have
543 uint64_t memusage = lzma_memusage(&strm);
544 uint64_t memlimit = hardware_memlimit_get();
546 // Round the memory limit down and usage up.
547 // This way we don't display a ridiculous
548 // message like "Limit was 9 MiB, but 9 MiB
549 // would have been needed".
550 memusage = (memusage + 1024 * 1024 - 1)
552 memlimit /= 1024 * 1024;
554 message_error(_("Limit was %s MiB, "
557 uint64_to_str(memlimit, 0),
558 uint64_to_str(memusage, 1));
565 // Show progress information under certain conditions.
566 message_progress_update();
573 /// Copy from input file to output file without processing the data in any
574 /// way. This is used only when trying to decompress unrecognized files
575 /// with --decompress --stdout --force, so the output is always stdout.
577 coder_passthru(file_pair *pair)
579 while (strm.avail_in != 0) {
583 if (io_write(pair, &in_buf, strm.avail_in))
586 strm.total_in += strm.avail_in;
587 strm.total_out = strm.total_in;
588 message_progress_update();
590 strm.avail_in = io_read(pair, &in_buf, IO_BUFFER_SIZE);
591 if (strm.avail_in == SIZE_MAX)
600 coder_run(const char *filename)
602 // Try to open the input and output files.
603 file_pair *pair = io_open(filename);
607 // Initialize the progress indicator.
608 const uint64_t in_size = pair->src_st.st_size <= (off_t)(0)
609 ? 0 : (uint64_t)(pair->src_st.st_size);
610 message_progress_start(&strm, pair->src_name, in_size);
612 // Assume that something goes wrong.
613 bool success = false;
615 // Read the first chunk of input data. This is needed to detect
616 // the input file type (for now, only for decompression).
617 strm.next_in = in_buf.u8;
618 strm.avail_in = io_read(pair, &in_buf, IO_BUFFER_SIZE);
620 switch (coder_init(pair)) {
621 case CODER_INIT_NORMAL:
622 success = coder_normal(pair);
625 case CODER_INIT_PASSTHRU:
626 success = coder_passthru(pair);
629 case CODER_INIT_ERROR:
633 message_progress_end(success);
635 // Close the file pair. It needs to know if coding was successful to
636 // know if the source or target file should be unlinked.
637 io_close(pair, success);