1 /* -*- mode: C; c-basic-offset: 4; indent-tabs-mode: nil -*- */
2 // vim: expandtab:ts=8:sw=4:softtabstop=4:
3 ///////////////////////////////////////////////////////////////////////////////
4 //
5 /// \file coder.c
6 /// \brief Compresses or uncompresses a file
7 //
8 // Author: Lasse Collin
9 //
10 // This file has been put into the public domain.
11 // You can do whatever you want with this file.
12 //
13 ///////////////////////////////////////////////////////////////////////////////
14
15 #include "private.h"
16
17
18 /// Return value type for coder_init().
19 enum coder_init_ret {
20 CODER_INIT_NORMAL,
21 CODER_INIT_PASSTHRU,
22 CODER_INIT_ERROR,
23 };
24
25
26 enum operation_mode opt_mode = MODE_COMPRESS;
27
28 enum format_type opt_format = FORMAT_AUTO;
29
30
31 /// Stream used to communicate with liblzma
32 static lzma_stream strm = LZMA_STREAM_INIT;
33
34 /// Filters needed for all encoding all formats, and also decoding in raw data
35 static lzma_filter filters[LZMA_FILTERS_MAX + 1];
36
37 /// Input and output buffers
38 static uint8_t in_buf[IO_BUFFER_SIZE];
39 static uint8_t out_buf[IO_BUFFER_SIZE];
40
41 /// Number of filters. Zero indicates that we are using a preset.
42 static size_t filters_count = 0;
43
44 /// Number of the preset (0-9)
45 static size_t preset_number = 6;
46
47 /// True if we should auto-adjust the compression settings to use less memory
48 /// if memory usage limit is too low for the original settings.
49 static bool auto_adjust = true;
50
51 /// Indicate if no preset has been explicitly given. In that case, if we need
52 /// to auto-adjust for lower memory usage, we won't print a warning.
53 static bool preset_default = true;
54
55 /// If a preset is used (no custom filter chain) and preset_extreme is true,
56 /// a significantly slower compression is used to achieve slightly better
57 /// compression ratio.
58 static bool preset_extreme = false;
59
60 /// Integrity check type
61 #ifdef HAVE_CHECK_CRC64
62 static lzma_check check = LZMA_CHECK_CRC64;
63 #else
64 static lzma_check check = LZMA_CHECK_CRC32;
65 #endif
66
67
68 extern void
coder_set_check(lzma_check new_check)69 coder_set_check(lzma_check new_check)
70 {
71 check = new_check;
72 return;
73 }
74
75
76 extern void
coder_set_preset(size_t new_preset)77 coder_set_preset(size_t new_preset)
78 {
79 preset_number = new_preset;
80 preset_default = false;
81 return;
82 }
83
84
85 extern void
coder_set_extreme(void)86 coder_set_extreme(void)
87 {
88 preset_extreme = true;
89 return;
90 }
91
92
93 extern void
coder_add_filter(lzma_vli id,void * options)94 coder_add_filter(lzma_vli id, void *options)
95 {
96 if (filters_count == LZMA_FILTERS_MAX)
97 message_fatal(_("Maximum number of filters is four"));
98
99 filters[filters_count].id = id;
100 filters[filters_count].options = options;
101 ++filters_count;
102
103 return;
104 }
105
106
107 static void lzma_attribute((noreturn))
memlimit_too_small(uint64_t memory_usage,uint64_t memory_limit)108 memlimit_too_small(uint64_t memory_usage, uint64_t memory_limit)
109 {
110 message_fatal(_("Memory usage limit (%" PRIu64 " MiB) is too small "
111 "for the given filter setup (%" PRIu64 " MiB)"),
112 memory_limit >> 20, memory_usage >> 20);
113 }
114
115
116 extern void
coder_set_compression_settings(void)117 coder_set_compression_settings(void)
118 {
119 // Options for LZMA1 or LZMA2 in case we are using a preset.
120 static lzma_options_lzma opt_lzma;
121
122 if (filters_count == 0) {
123 // We are using a preset. This is not a good idea in raw mode
124 // except when playing around with things. Different versions
125 // of this software may use different options in presets, and
126 // thus make uncompressing the raw data difficult.
127 if (opt_format == FORMAT_RAW) {
128 // The message is shown only if warnings are allowed
129 // but the exit status isn't changed.
130 message(V_WARNING, _("Using a preset in raw mode "
131 "is discouraged."));
132 message(V_WARNING, _("The exact options of the "
133 "presets may vary between software "
134 "versions."));
135 }
136
137 // Get the preset for LZMA1 or LZMA2.
138 if (preset_extreme)
139 preset_number |= LZMA_PRESET_EXTREME;
140
141 if (lzma_lzma_preset(&opt_lzma, preset_number))
142 message_bug();
143
144 // Use LZMA2 except with --format=lzma we use LZMA1.
145 filters[0].id = opt_format == FORMAT_LZMA
146 ? LZMA_FILTER_LZMA1 : LZMA_FILTER_LZMA2;
147 filters[0].options = &opt_lzma;
148 filters_count = 1;
149 } else {
150 preset_default = false;
151 }
152
153 // Terminate the filter options array.
154 filters[filters_count].id = LZMA_VLI_UNKNOWN;
155
156 // If we are using the .lzma format, allow exactly one filter
157 // which has to be LZMA1.
158 if (opt_format == FORMAT_LZMA && (filters_count != 1
159 || filters[0].id != LZMA_FILTER_LZMA1))
160 message_fatal(_("The .lzma format supports only "
161 "the LZMA1 filter"));
162
163 // If we are using the .xz format, make sure that there is no LZMA1
164 // filter to prevent LZMA_PROG_ERROR.
165 if (opt_format == FORMAT_XZ)
166 for (size_t i = 0; i < filters_count; ++i)
167 if (filters[i].id == LZMA_FILTER_LZMA1)
168 message_fatal(_("LZMA1 cannot be used "
169 "with the .xz format"));
170
171 // Print the selected filter chain.
172 message_filters(V_DEBUG, filters);
173
174 // If using --format=raw, we can be decoding. The memusage function
175 // also validates the filter chain and the options used for the
176 // filters.
177 const uint64_t memory_limit = hardware_memlimit_get();
178 uint64_t memory_usage;
179 if (opt_mode == MODE_COMPRESS)
180 memory_usage = lzma_raw_encoder_memusage(filters);
181 else
182 memory_usage = lzma_raw_decoder_memusage(filters);
183
184 if (memory_usage == UINT64_MAX)
185 message_fatal("Unsupported filter chain or filter options");
186
187 // Print memory usage info.
188 message(V_DEBUG, _("%s MiB (%s B) of memory is required per thread, "
189 "limit is %s MiB (%s B)"),
190 uint64_to_str(memory_usage >> 20, 0),
191 uint64_to_str(memory_usage, 1),
192 uint64_to_str(memory_limit >> 20, 2),
193 uint64_to_str(memory_limit, 3));
194
195 if (memory_usage > memory_limit) {
196 // If --no-auto-adjust was used or we didn't find LZMA1 or
197 // LZMA2 as the last filter, give an error immediatelly.
198 // --format=raw implies --no-auto-adjust.
199 if (!auto_adjust || opt_format == FORMAT_RAW)
200 memlimit_too_small(memory_usage, memory_limit);
201
202 assert(opt_mode == MODE_COMPRESS);
203
204 // Look for the last filter if it is LZMA2 or LZMA1, so
205 // we can make it use less RAM. With other filters we don't
206 // know what to do.
207 size_t i = 0;
208 while (filters[i].id != LZMA_FILTER_LZMA2
209 && filters[i].id != LZMA_FILTER_LZMA1) {
210 if (filters[i].id == LZMA_VLI_UNKNOWN)
211 memlimit_too_small(memory_usage, memory_limit);
212
213 ++i;
214 }
215
216 // Decrease the dictionary size until we meet the memory
217 // usage limit. First round down to full mebibytes.
218 lzma_options_lzma *opt = filters[i].options;
219 const uint32_t orig_dict_size = opt->dict_size;
220 opt->dict_size &= ~((UINT32_C(1) << 20) - 1);
221 while (true) {
222 // If it is below 1 MiB, auto-adjusting failed. We
223 // could be more sophisticated and scale it down even
224 // more, but let's see if many complain about this
225 // version.
226 //
227 // FIXME: Displays the scaled memory usage instead
228 // of the original.
229 if (opt->dict_size < (UINT32_C(1) << 20))
230 memlimit_too_small(memory_usage, memory_limit);
231
232 memory_usage = lzma_raw_encoder_memusage(filters);
233 if (memory_usage == UINT64_MAX)
234 message_bug();
235
236 // Accept it if it is low enough.
237 if (memory_usage <= memory_limit)
238 break;
239
240 // Otherwise 1 MiB down and try again. I hope this
241 // isn't too slow method for cases where the original
242 // dict_size is very big.
243 opt->dict_size -= UINT32_C(1) << 20;
244 }
245
246 // Tell the user that we decreased the dictionary size.
247 // However, omit the message if no preset or custom chain
248 // was given. FIXME: Always warn?
249 if (!preset_default)
250 message(V_WARNING, "Adjusted LZMA%c dictionary size "
251 "from %s MiB to %s MiB to not exceed "
252 "the memory usage limit of %s MiB",
253 filters[i].id == LZMA_FILTER_LZMA2
254 ? '2' : '1',
255 uint64_to_str(orig_dict_size >> 20, 0),
256 uint64_to_str(opt->dict_size >> 20, 1),
257 uint64_to_str(memory_limit >> 20, 2));
258 }
259
260 /*
261 // Limit the number of worker threads so that memory usage
262 // limit isn't exceeded.
263 assert(memory_usage > 0);
264 size_t thread_limit = memory_limit / memory_usage;
265 if (thread_limit == 0)
266 thread_limit = 1;
267
268 if (opt_threads > thread_limit)
269 opt_threads = thread_limit;
270 */
271
272 return;
273 }
274
275
276 /// Return true if the data in in_buf seems to be in the .xz format.
277 static bool
is_format_xz(void)278 is_format_xz(void)
279 {
280 return strm.avail_in >= 6 && memcmp(in_buf, "\3757zXZ", 6) == 0;
281 }
282
283
284 /// Return true if the data in in_buf seems to be in the .lzma format.
285 static bool
is_format_lzma(void)286 is_format_lzma(void)
287 {
288 // The .lzma header is 13 bytes.
289 if (strm.avail_in < 13)
290 return false;
291
292 // Decode the LZMA1 properties.
293 lzma_filter filter = { .id = LZMA_FILTER_LZMA1 };
294 if (lzma_properties_decode(&filter, NULL, in_buf, 5) != LZMA_OK)
295 return false;
296
297 // A hack to ditch tons of false positives: We allow only dictionary
298 // sizes that are 2^n or 2^n + 2^(n-1) or UINT32_MAX. LZMA_Alone
299 // created only files with 2^n, but accepts any dictionary size.
300 // If someone complains, this will be reconsidered.
301 lzma_options_lzma *opt = filter.options;
302 const uint32_t dict_size = opt->dict_size;
303 free(opt);
304
305 if (dict_size != UINT32_MAX) {
306 uint32_t d = dict_size - 1;
307 d |= d >> 2;
308 d |= d >> 3;
309 d |= d >> 4;
310 d |= d >> 8;
311 d |= d >> 16;
312 ++d;
313 if (d != dict_size || dict_size == 0)
314 return false;
315 }
316
317 // Another hack to ditch false positives: Assume that if the
318 // uncompressed size is known, it must be less than 256 GiB.
319 // Again, if someone complains, this will be reconsidered.
320 uint64_t uncompressed_size = 0;
321 for (size_t i = 0; i < 8; ++i)
322 uncompressed_size |= (uint64_t)(in_buf[5 + i]) << (i * 8);
323
324 if (uncompressed_size != UINT64_MAX
325 && uncompressed_size > (UINT64_C(1) << 38))
326 return false;
327
328 return true;
329 }
330
331
332 /// Detect the input file type (for now, this done only when decompressing),
333 /// and initialize an appropriate coder. Return value indicates if a normal
334 /// liblzma-based coder was initialized (CODER_INIT_NORMAL), if passthru
335 /// mode should be used (CODER_INIT_PASSTHRU), or if an error occurred
336 /// (CODER_INIT_ERROR).
337 static enum coder_init_ret
coder_init(file_pair * pair)338 coder_init(file_pair *pair)
339 {
340 lzma_ret ret = LZMA_PROG_ERROR;
341
342 if (opt_mode == MODE_COMPRESS) {
343 switch (opt_format) {
344 case FORMAT_AUTO:
345 // args.c ensures this.
346 assert(0);
347 break;
348
349 case FORMAT_XZ:
350 ret = lzma_stream_encoder(&strm, filters, check);
351 break;
352
353 case FORMAT_LZMA:
354 ret = lzma_alone_encoder(&strm, filters[0].options);
355 break;
356
357 case FORMAT_RAW:
358 ret = lzma_raw_encoder(&strm, filters);
359 break;
360 }
361 } else {
362 const uint32_t flags = LZMA_TELL_UNSUPPORTED_CHECK
363 | LZMA_CONCATENATED;
364
365 // We abuse FORMAT_AUTO to indicate unknown file format,
366 // for which we may consider passthru mode.
367 enum format_type init_format = FORMAT_AUTO;
368
369 switch (opt_format) {
370 case FORMAT_AUTO:
371 if (is_format_xz())
372 init_format = FORMAT_XZ;
373 else if (is_format_lzma())
374 init_format = FORMAT_LZMA;
375 break;
376
377 case FORMAT_XZ:
378 if (is_format_xz())
379 init_format = FORMAT_XZ;
380 break;
381
382 case FORMAT_LZMA:
383 if (is_format_lzma())
384 init_format = FORMAT_LZMA;
385 break;
386
387 case FORMAT_RAW:
388 init_format = FORMAT_RAW;
389 break;
390 }
391
392 switch (init_format) {
393 case FORMAT_AUTO:
394 // Uknown file format. If --decompress --stdout
395 // --force have been given, then we copy the input
396 // as is to stdout. Checking for MODE_DECOMPRESS
397 // is needed, because we don't want to do use
398 // passthru mode with --test.
399 if (opt_mode == MODE_DECOMPRESS
400 && opt_stdout && opt_force)
401 return CODER_INIT_PASSTHRU;
402
403 ret = LZMA_FORMAT_ERROR;
404 break;
405
406 case FORMAT_XZ:
407 ret = lzma_stream_decoder(&strm,
408 hardware_memlimit_get(), flags);
409 break;
410
411 case FORMAT_LZMA:
412 ret = lzma_alone_decoder(&strm,
413 hardware_memlimit_get());
414 break;
415
416 case FORMAT_RAW:
417 // Memory usage has already been checked in
418 // coder_set_compression_settings().
419 ret = lzma_raw_decoder(&strm, filters);
420 break;
421 }
422 }
423
424 if (ret != LZMA_OK) {
425 message_error("%s: %s", pair->src_name, message_strm(ret));
426 return CODER_INIT_ERROR;
427 }
428
429 return CODER_INIT_NORMAL;
430 }
431
432
433 /// Compress or decompress using liblzma.
434 static bool
coder_normal(file_pair * pair)435 coder_normal(file_pair *pair)
436 {
437 // Encoder needs to know when we have given all the input to it.
438 // The decoders need to know it too when we are using
439 // LZMA_CONCATENATED. We need to check for src_eof here, because
440 // the first input chunk has been already read, and that may
441 // have been the only chunk we will read.
442 lzma_action action = pair->src_eof ? LZMA_FINISH : LZMA_RUN;
443
444 lzma_ret ret;
445
446 // Assume that something goes wrong.
447 bool success = false;
448
449 strm.next_out = out_buf;
450 strm.avail_out = IO_BUFFER_SIZE;
451
452 while (!user_abort) {
453 // Fill the input buffer if it is empty and we haven't reached
454 // end of file yet.
455 if (strm.avail_in == 0 && !pair->src_eof) {
456 strm.next_in = in_buf;
457 strm.avail_in = io_read(pair, in_buf, IO_BUFFER_SIZE);
458
459 if (strm.avail_in == SIZE_MAX)
460 break;
461
462 if (pair->src_eof)
463 action = LZMA_FINISH;
464 }
465
466 // Let liblzma do the actual work.
467 ret = lzma_code(&strm, action);
468
469 // Write out if the output buffer became full.
470 if (strm.avail_out == 0) {
471 if (opt_mode != MODE_TEST && io_write(pair, out_buf,
472 IO_BUFFER_SIZE - strm.avail_out))
473 break;
474
475 strm.next_out = out_buf;
476 strm.avail_out = IO_BUFFER_SIZE;
477 }
478
479 if (ret != LZMA_OK) {
480 // Determine if the return value indicates that we
481 // won't continue coding.
482 const bool stop = ret != LZMA_NO_CHECK
483 && ret != LZMA_UNSUPPORTED_CHECK;
484
485 if (stop) {
486 // Write the remaining bytes even if something
487 // went wrong, because that way the user gets
488 // as much data as possible, which can be good
489 // when trying to get at least some useful
490 // data out of damaged files.
491 if (opt_mode != MODE_TEST && io_write(pair,
492 out_buf, IO_BUFFER_SIZE
493 - strm.avail_out))
494 break;
495 }
496
497 if (ret == LZMA_STREAM_END) {
498 // Check that there is no trailing garbage.
499 // This is needed for LZMA_Alone and raw
500 // streams.
501 if (strm.avail_in == 0 && !pair->src_eof) {
502 // Try reading one more byte.
503 // Hopefully we don't get any more
504 // input, and thus pair->src_eof
505 // becomes true.
506 strm.avail_in = io_read(
507 pair, in_buf, 1);
508 if (strm.avail_in == SIZE_MAX)
509 break;
510
511 assert(strm.avail_in == 0
512 || strm.avail_in == 1);
513 }
514
515 if (strm.avail_in == 0) {
516 assert(pair->src_eof);
517 success = true;
518 break;
519 }
520
521 // We hadn't reached the end of the file.
522 ret = LZMA_DATA_ERROR;
523 assert(stop);
524 }
525
526 // If we get here and stop is true, something went
527 // wrong and we print an error. Otherwise it's just
528 // a warning and coding can continue.
529 if (stop) {
530 message_error("%s: %s", pair->src_name,
531 message_strm(ret));
532 } else {
533 message_warning("%s: %s", pair->src_name,
534 message_strm(ret));
535
536 // When compressing, all possible errors set
537 // stop to true.
538 assert(opt_mode != MODE_COMPRESS);
539 }
540
541 if (ret == LZMA_MEMLIMIT_ERROR) {
542 // Figure out how much memory it would have
543 // actually needed.
544 uint64_t memusage = lzma_memusage(&strm);
545 uint64_t memlimit = hardware_memlimit_get();
546
547 // Round the memory limit down and usage up.
548 // This way we don't display a ridiculous
549 // message like "Limit was 9 MiB, but 9 MiB
550 // would have been needed".
551 memusage = (memusage + 1024 * 1024 - 1)
552 / (1024 * 1024);
553 memlimit /= 1024 * 1024;
554
555 message_error(_("Limit was %s MiB, "
556 "but %s MiB would "
557 "have been needed"),
558 uint64_to_str(memlimit, 0),
559 uint64_to_str(memusage, 1));
560 }
561
562 if (stop)
563 break;
564 }
565
566 // Show progress information under certain conditions.
567 message_progress_update();
568 }
569
570 return success;
571 }
572
573
574 /// Copy from input file to output file without processing the data in any
575 /// way. This is used only when trying to decompress unrecognized files
576 /// with --decompress --stdout --force, so the output is always stdout.
577 static bool
coder_passthru(file_pair * pair)578 coder_passthru(file_pair *pair)
579 {
580 while (strm.avail_in != 0) {
581 if (user_abort)
582 return false;
583
584 if (io_write(pair, in_buf, strm.avail_in))
585 return false;
586
587 strm.total_in += strm.avail_in;
588 strm.total_out = strm.total_in;
589 message_progress_update();
590
591 strm.avail_in = io_read(pair, in_buf, IO_BUFFER_SIZE);
592 if (strm.avail_in == SIZE_MAX)
593 return false;
594 }
595
596 return true;
597 }
598
599
600 extern void
coder_run(const char * filename)601 coder_run(const char *filename)
602 {
603 // Try to open the input and output files.
604 file_pair *pair = io_open(filename);
605 if (pair == NULL)
606 return;
607
608 // Initialize the progress indicator.
609 const uint64_t in_size = pair->src_st.st_size <= (off_t)(0)
610 ? 0 : (uint64_t)(pair->src_st.st_size);
611 message_progress_start(&strm, pair->src_name, in_size);
612
613 // Assume that something goes wrong.
614 bool success = false;
615
616 // Read the first chunk of input data. This is needed to detect
617 // the input file type (for now, only for decompression).
618 strm.next_in = in_buf;
619 strm.avail_in = io_read(pair, in_buf, IO_BUFFER_SIZE);
620
621 switch (coder_init(pair)) {
622 case CODER_INIT_NORMAL:
623 success = coder_normal(pair);
624 break;
625
626 case CODER_INIT_PASSTHRU:
627 success = coder_passthru(pair);
628 break;
629
630 case CODER_INIT_ERROR:
631 break;
632 }
633
634 message_progress_end(success);
635
636 // Close the file pair. It needs to know if coding was successful to
637 // know if the source or target file should be unlinked.
638 io_close(pair, success);
639
640 return;
641 }
642