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
36 static uint8_t in_buf[IO_BUFFER_SIZE];
37 static uint8_t out_buf[IO_BUFFER_SIZE];
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_Alone format, allow exactly one filter
155 // which has to be LZMA.
156 if (opt_format == FORMAT_LZMA && (filters_count != 1
157 || filters[0].id != LZMA_FILTER_LZMA1))
158 message_fatal(_("With --format=lzma only the LZMA1 filter "
161 // Print the selected filter chain.
162 message_filters(V_DEBUG, filters);
164 // If using --format=raw, we can be decoding. The memusage function
165 // also validates the filter chain and the options used for the
167 const uint64_t memory_limit = hardware_memlimit_get();
168 uint64_t memory_usage;
169 if (opt_mode == MODE_COMPRESS)
170 memory_usage = lzma_raw_encoder_memusage(filters);
172 memory_usage = lzma_raw_decoder_memusage(filters);
174 if (memory_usage == UINT64_MAX)
175 message_fatal("Unsupported filter chain or filter options");
177 // Print memory usage info.
178 message(V_DEBUG, _("%s MiB (%s B) of memory is required per thread, "
179 "limit is %s MiB (%s B)"),
180 uint64_to_str(memory_usage >> 20, 0),
181 uint64_to_str(memory_usage, 1),
182 uint64_to_str(memory_limit >> 20, 2),
183 uint64_to_str(memory_limit, 3));
185 if (memory_usage > memory_limit) {
186 // If --no-auto-adjust was used or we didn't find LZMA1 or
187 // LZMA2 as the last filter, give an error immediatelly.
188 // --format=raw implies --no-auto-adjust.
189 if (!auto_adjust || opt_format == FORMAT_RAW)
190 memlimit_too_small(memory_usage, memory_limit);
192 assert(opt_mode == MODE_COMPRESS);
194 // Look for the last filter if it is LZMA2 or LZMA1, so
195 // we can make it use less RAM. With other filters we don't
198 while (filters[i].id != LZMA_FILTER_LZMA2
199 && filters[i].id != LZMA_FILTER_LZMA1) {
200 if (filters[i].id == LZMA_VLI_UNKNOWN)
201 memlimit_too_small(memory_usage, memory_limit);
206 // Decrease the dictionary size until we meet the memory
207 // usage limit. First round down to full mebibytes.
208 lzma_options_lzma *opt = filters[i].options;
209 const uint32_t orig_dict_size = opt->dict_size;
210 opt->dict_size &= ~((UINT32_C(1) << 20) - 1);
212 // If it is below 1 MiB, auto-adjusting failed. We
213 // could be more sophisticated and scale it down even
214 // more, but let's see if many complain about this
217 // FIXME: Displays the scaled memory usage instead
219 if (opt->dict_size < (UINT32_C(1) << 20))
220 memlimit_too_small(memory_usage, memory_limit);
222 memory_usage = lzma_raw_encoder_memusage(filters);
223 if (memory_usage == UINT64_MAX)
226 // Accept it if it is low enough.
227 if (memory_usage <= memory_limit)
230 // Otherwise 1 MiB down and try again. I hope this
231 // isn't too slow method for cases where the original
232 // dict_size is very big.
233 opt->dict_size -= UINT32_C(1) << 20;
236 // Tell the user that we decreased the dictionary size.
237 // However, omit the message if no preset or custom chain
238 // was given. FIXME: Always warn?
240 message(V_WARNING, "Adjusted LZMA%c dictionary size "
241 "from %s MiB to %s MiB to not exceed "
242 "the memory usage limit of %s MiB",
243 filters[i].id == LZMA_FILTER_LZMA2
245 uint64_to_str(orig_dict_size >> 20, 0),
246 uint64_to_str(opt->dict_size >> 20, 1),
247 uint64_to_str(memory_limit >> 20, 2));
251 // Limit the number of worker threads so that memory usage
252 // limit isn't exceeded.
253 assert(memory_usage > 0);
254 size_t thread_limit = memory_limit / memory_usage;
255 if (thread_limit == 0)
258 if (opt_threads > thread_limit)
259 opt_threads = thread_limit;
266 /// Return true if the data in in_buf seems to be in the .xz format.
270 return strm.avail_in >= 6 && memcmp(in_buf, "\3757zXZ", 6) == 0;
274 /// Return true if the data in in_buf seems to be in the .lzma format.
278 // The .lzma header is 13 bytes.
279 if (strm.avail_in < 13)
282 // Decode the LZMA1 properties.
283 lzma_filter filter = { .id = LZMA_FILTER_LZMA1 };
284 if (lzma_properties_decode(&filter, NULL, in_buf, 5) != LZMA_OK)
287 // A hack to ditch tons of false positives: We allow only dictionary
288 // sizes that are 2^n or 2^n + 2^(n-1) or UINT32_MAX. LZMA_Alone
289 // created only files with 2^n, but accepts any dictionary size.
290 // If someone complains, this will be reconsidered.
291 lzma_options_lzma *opt = filter.options;
292 const uint32_t dict_size = opt->dict_size;
295 if (dict_size != UINT32_MAX) {
296 uint32_t d = dict_size - 1;
303 if (d != dict_size || dict_size == 0)
307 // Another hack to ditch false positives: Assume that if the
308 // uncompressed size is known, it must be less than 256 GiB.
309 // Again, if someone complains, this will be reconsidered.
310 uint64_t uncompressed_size = 0;
311 for (size_t i = 0; i < 8; ++i)
312 uncompressed_size |= (uint64_t)(in_buf[5 + i]) << (i * 8);
314 if (uncompressed_size != UINT64_MAX
315 && uncompressed_size > (UINT64_C(1) << 38))
322 /// Detect the input file type (for now, this done only when decompressing),
323 /// and initialize an appropriate coder. Return value indicates if a normal
324 /// liblzma-based coder was initialized (CODER_INIT_NORMAL), if passthru
325 /// mode should be used (CODER_INIT_PASSTHRU), or if an error occurred
326 /// (CODER_INIT_ERROR).
327 static enum coder_init_ret
328 coder_init(file_pair *pair)
330 lzma_ret ret = LZMA_PROG_ERROR;
332 if (opt_mode == MODE_COMPRESS) {
333 switch (opt_format) {
335 // args.c ensures this.
340 ret = lzma_stream_encoder(&strm, filters, check);
344 ret = lzma_alone_encoder(&strm, filters[0].options);
348 ret = lzma_raw_encoder(&strm, filters);
352 const uint32_t flags = LZMA_TELL_UNSUPPORTED_CHECK
355 // We abuse FORMAT_AUTO to indicate unknown file format,
356 // for which we may consider passthru mode.
357 enum format_type init_format = FORMAT_AUTO;
359 switch (opt_format) {
362 init_format = FORMAT_XZ;
363 else if (is_format_lzma())
364 init_format = FORMAT_LZMA;
369 init_format = FORMAT_XZ;
373 if (is_format_lzma())
374 init_format = FORMAT_LZMA;
378 init_format = FORMAT_RAW;
382 switch (init_format) {
384 // Uknown file format. If --decompress --stdout
385 // --force have been given, then we copy the input
386 // as is to stdout. Checking for MODE_DECOMPRESS
387 // is needed, because we don't want to do use
388 // passthru mode with --test.
389 if (opt_mode == MODE_DECOMPRESS
390 && opt_stdout && opt_force)
391 return CODER_INIT_PASSTHRU;
393 ret = LZMA_FORMAT_ERROR;
397 ret = lzma_stream_decoder(&strm,
398 hardware_memlimit_get(), flags);
402 ret = lzma_alone_decoder(&strm,
403 hardware_memlimit_get());
407 // Memory usage has already been checked in
408 // coder_set_compression_settings().
409 ret = lzma_raw_decoder(&strm, filters);
414 if (ret != LZMA_OK) {
415 message_error("%s: %s", pair->src_name, message_strm(ret));
416 return CODER_INIT_ERROR;
419 return CODER_INIT_NORMAL;
423 /// Compress or decompress using liblzma.
425 coder_normal(file_pair *pair)
427 // Encoder needs to know when we have given all the input to it.
428 // The decoders need to know it too when we are using
429 // LZMA_CONCATENATED. We need to check for src_eof here, because
430 // the first input chunk has been already read, and that may
431 // have been the only chunk we will read.
432 lzma_action action = pair->src_eof ? LZMA_FINISH : LZMA_RUN;
436 // Assume that something goes wrong.
437 bool success = false;
439 strm.next_out = out_buf;
440 strm.avail_out = IO_BUFFER_SIZE;
442 while (!user_abort) {
443 // Fill the input buffer if it is empty and we haven't reached
445 if (strm.avail_in == 0 && !pair->src_eof) {
446 strm.next_in = in_buf;
447 strm.avail_in = io_read(pair, in_buf, IO_BUFFER_SIZE);
449 if (strm.avail_in == SIZE_MAX)
453 action = LZMA_FINISH;
456 // Let liblzma do the actual work.
457 ret = lzma_code(&strm, action);
459 // Write out if the output buffer became full.
460 if (strm.avail_out == 0) {
461 if (opt_mode != MODE_TEST && io_write(pair, out_buf,
462 IO_BUFFER_SIZE - strm.avail_out))
465 strm.next_out = out_buf;
466 strm.avail_out = IO_BUFFER_SIZE;
469 if (ret != LZMA_OK) {
470 // Determine if the return value indicates that we
471 // won't continue coding.
472 const bool stop = ret != LZMA_NO_CHECK
473 && ret != LZMA_UNSUPPORTED_CHECK;
476 // Write the remaining bytes even if something
477 // went wrong, because that way the user gets
478 // as much data as possible, which can be good
479 // when trying to get at least some useful
480 // data out of damaged files.
481 if (opt_mode != MODE_TEST && io_write(pair,
482 out_buf, IO_BUFFER_SIZE
487 if (ret == LZMA_STREAM_END) {
488 // Check that there is no trailing garbage.
489 // This is needed for LZMA_Alone and raw
491 if (strm.avail_in == 0 && !pair->src_eof) {
492 // Try reading one more byte.
493 // Hopefully we don't get any more
494 // input, and thus pair->src_eof
496 strm.avail_in = io_read(
498 if (strm.avail_in == SIZE_MAX)
501 assert(strm.avail_in == 0
502 || strm.avail_in == 1);
505 if (strm.avail_in == 0) {
506 assert(pair->src_eof);
511 // We hadn't reached the end of the file.
512 ret = LZMA_DATA_ERROR;
516 // If we get here and stop is true, something went
517 // wrong and we print an error. Otherwise it's just
518 // a warning and coding can continue.
520 message_error("%s: %s", pair->src_name,
523 message_warning("%s: %s", pair->src_name,
526 // When compressing, all possible errors set
528 assert(opt_mode != MODE_COMPRESS);
531 if (ret == LZMA_MEMLIMIT_ERROR) {
532 // Figure out how much memory it would have
534 uint64_t memusage = lzma_memusage(&strm);
535 uint64_t memlimit = hardware_memlimit_get();
537 // Round the memory limit down and usage up.
538 // This way we don't display a ridiculous
539 // message like "Limit was 9 MiB, but 9 MiB
540 // would have been needed".
541 memusage = (memusage + 1024 * 1024 - 1)
543 memlimit /= 1024 * 1024;
545 message_error(_("Limit was %s MiB, "
548 uint64_to_str(memlimit, 0),
549 uint64_to_str(memusage, 1));
556 // Show progress information under certain conditions.
557 message_progress_update();
564 /// Copy from input file to output file without processing the data in any
565 /// way. This is used only when trying to decompress unrecognized files
566 /// with --decompress --stdout --force, so the output is always stdout.
568 coder_passthru(file_pair *pair)
570 while (strm.avail_in != 0) {
574 if (io_write(pair, in_buf, strm.avail_in))
577 strm.total_in += strm.avail_in;
578 strm.total_out = strm.total_in;
579 message_progress_update();
581 strm.avail_in = io_read(pair, in_buf, IO_BUFFER_SIZE);
582 if (strm.avail_in == SIZE_MAX)
591 coder_run(const char *filename)
593 // Try to open the input and output files.
594 file_pair *pair = io_open(filename);
598 // Initialize the progress indicator.
599 const uint64_t in_size = pair->src_st.st_size <= (off_t)(0)
600 ? 0 : (uint64_t)(pair->src_st.st_size);
601 message_progress_start(&strm, pair->src_name, in_size);
603 // Assume that something goes wrong.
604 bool success = false;
606 // Read the first chunk of input data. This is needed to detect
607 // the input file type (for now, only for decompression).
608 strm.next_in = in_buf;
609 strm.avail_in = io_read(pair, in_buf, IO_BUFFER_SIZE);
611 switch (coder_init(pair)) {
612 case CODER_INIT_NORMAL:
613 success = coder_normal(pair);
616 case CODER_INIT_PASSTHRU:
617 success = coder_passthru(pair);
620 case CODER_INIT_ERROR:
624 message_progress_end(success);
626 // Close the file pair. It needs to know if coding was successful to
627 // know if the source or target file should be unlinked.
628 io_close(pair, success);