1 /* Small bzip2 deflate implementation, by Rob Landley (rob@landley.net).
3 Based on bzip2 decompression code by Julian R Seward (jseward@acm.org),
4 which also acknowledges contributions by Mike Burrows, David Wheeler,
5 Peter Fenwick, Alistair Moffat, Radford Neal, Ian H. Witten,
6 Robert Sedgewick, and Jon L. Bentley.
8 This code is licensed under the LGPLv2:
9 LGPL (http://www.gnu.org/copyleft/lgpl.html
13 Size and speed optimizations by Manuel Novoa III (mjn3@codepoet.org).
15 More efficient reading of Huffman codes, a streamlined read_bunzip()
16 function, and various other tweaks. In (limited) tests, approximately
17 20% faster than bzcat on x86 and about 10% faster on arm.
19 Note that about 2/3 of the time is spent in read_unzip() reversing
20 the Burrows-Wheeler transformation. Much of that time is delay
21 resulting from cache misses.
23 I would ask that anyone benefiting from this work, especially those
24 using it in commercial products, consider making a donation to my local
25 non-profit hospice organization in the name of the woman I loved, who
26 passed away Feb. 12, 2003.
28 In memory of Toni W. Hagan
30 Hospice of Acadiana, Inc.
31 2600 Johnston St., Suite 200
32 Lafayette, LA 70503-3240
34 Phone (337) 232-1234 or 1-800-738-2226
37 https://www.hospiceacadiana.com/
43 Made it fit for running in Linux Kernel by Alain Knaff (alain@knaff.lu)
50 #include <linux/decompress/bunzip2.h>
53 #include <linux/decompress/mm.h>
54 #include <linux/crc32poly.h>
57 #define INT_MAX 0x7fffffff
60 /* Constants for Huffman coding */
62 #define GROUP_SIZE 50 /* 64 would have been more efficient */
63 #define MAX_HUFCODE_BITS 20 /* Longest Huffman code allowed */
64 #define MAX_SYMBOLS 258 /* 256 literals + RUNA + RUNB */
68 /* Status return values */
70 #define RETVAL_LAST_BLOCK (-1)
71 #define RETVAL_NOT_BZIP_DATA (-2)
72 #define RETVAL_UNEXPECTED_INPUT_EOF (-3)
73 #define RETVAL_UNEXPECTED_OUTPUT_EOF (-4)
74 #define RETVAL_DATA_ERROR (-5)
75 #define RETVAL_OUT_OF_MEMORY (-6)
76 #define RETVAL_OBSOLETE_INPUT (-7)
78 /* Other housekeeping constants */
79 #define BZIP2_IOBUF_SIZE 4096
81 /* This is what we know about each Huffman coding group */
83 /* We have an extra slot at the end of limit[] for a sentinal value. */
84 int limit[MAX_HUFCODE_BITS+1];
85 int base[MAX_HUFCODE_BITS];
86 int permute[MAX_SYMBOLS];
90 /* Structure holding all the housekeeping data, including IO buffers and
91 memory that persists between calls to bunzip */
93 /* State for interrupting output loop */
94 int writeCopies, writePos, writeRunCountdown, writeCount, writeCurrent;
95 /* I/O tracking data (file handles, buffers, positions, etc.) */
96 long (*fill)(void*, unsigned long);
97 long inbufCount, inbufPos /*, outbufPos*/;
98 unsigned char *inbuf /*,*outbuf*/;
99 unsigned int inbufBitCount, inbufBits;
100 /* The CRC values stored in the block header and calculated from the
102 unsigned int crc32Table[256], headerCRC, totalCRC, writeCRC;
103 /* Intermediate buffer and its size (in bytes) */
104 unsigned int *dbuf, dbufSize;
105 /* These things are a bit too big to go on the stack */
106 unsigned char selectors[32768]; /* nSelectors = 15 bits */
107 struct group_data groups[MAX_GROUPS]; /* Huffman coding tables */
108 int io_error; /* non-zero if we have IO error */
110 unsigned char symToByte[256], mtfSymbol[256];
114 /* Return the next nnn bits of input. All reads from the compressed input
115 are done through this function. All reads are big endian */
116 static unsigned int INIT get_bits(struct bunzip_data *bd, char bits_wanted)
118 unsigned int bits = 0;
120 /* If we need to get more data from the byte buffer, do so.
121 (Loop getting one byte at a time to enforce endianness and avoid
122 unaligned access.) */
123 while (bd->inbufBitCount < bits_wanted) {
124 /* If we need to read more data from file into byte buffer, do
126 if (bd->inbufPos == bd->inbufCount) {
129 bd->inbufCount = bd->fill(bd->inbuf, BZIP2_IOBUF_SIZE);
130 if (bd->inbufCount <= 0) {
131 bd->io_error = RETVAL_UNEXPECTED_INPUT_EOF;
136 /* Avoid 32-bit overflow (dump bit buffer to top of output) */
137 if (bd->inbufBitCount >= 24) {
138 bits = bd->inbufBits&((1 << bd->inbufBitCount)-1);
139 bits_wanted -= bd->inbufBitCount;
140 bits <<= bits_wanted;
141 bd->inbufBitCount = 0;
143 /* Grab next 8 bits of input from buffer. */
144 bd->inbufBits = (bd->inbufBits << 8)|bd->inbuf[bd->inbufPos++];
145 bd->inbufBitCount += 8;
147 /* Calculate result */
148 bd->inbufBitCount -= bits_wanted;
149 bits |= (bd->inbufBits >> bd->inbufBitCount)&((1 << bits_wanted)-1);
154 /* Unpacks the next block and sets up for the inverse burrows-wheeler step. */
156 static int INIT get_next_block(struct bunzip_data *bd)
158 struct group_data *hufGroup = NULL;
161 int dbufCount, nextSym, dbufSize, groupCount, selector,
162 i, j, k, t, runPos, symCount, symTotal, nSelectors, *byteCount;
163 unsigned char uc, *symToByte, *mtfSymbol, *selectors;
164 unsigned int *dbuf, origPtr;
167 dbufSize = bd->dbufSize;
168 selectors = bd->selectors;
169 byteCount = bd->byteCount;
170 symToByte = bd->symToByte;
171 mtfSymbol = bd->mtfSymbol;
173 /* Read in header signature and CRC, then validate signature.
174 (last block signature means CRC is for whole file, return now) */
175 i = get_bits(bd, 24);
176 j = get_bits(bd, 24);
177 bd->headerCRC = get_bits(bd, 32);
178 if ((i == 0x177245) && (j == 0x385090))
179 return RETVAL_LAST_BLOCK;
180 if ((i != 0x314159) || (j != 0x265359))
181 return RETVAL_NOT_BZIP_DATA;
182 /* We can add support for blockRandomised if anybody complains.
183 There was some code for this in busybox 1.0.0-pre3, but nobody ever
184 noticed that it didn't actually work. */
186 return RETVAL_OBSOLETE_INPUT;
187 origPtr = get_bits(bd, 24);
188 if (origPtr >= dbufSize)
189 return RETVAL_DATA_ERROR;
190 /* mapping table: if some byte values are never used (encoding things
191 like ascii text), the compression code removes the gaps to have fewer
192 symbols to deal with, and writes a sparse bitfield indicating which
193 values were present. We make a translation table to convert the
194 symbols back to the corresponding bytes. */
195 t = get_bits(bd, 16);
197 for (i = 0; i < 16; i++) {
198 if (t&(1 << (15-i))) {
199 k = get_bits(bd, 16);
200 for (j = 0; j < 16; j++)
202 symToByte[symTotal++] = (16*i)+j;
205 /* How many different Huffman coding groups does this block use? */
206 groupCount = get_bits(bd, 3);
207 if (groupCount < 2 || groupCount > MAX_GROUPS)
208 return RETVAL_DATA_ERROR;
209 /* nSelectors: Every GROUP_SIZE many symbols we select a new
210 Huffman coding group. Read in the group selector list,
211 which is stored as MTF encoded bit runs. (MTF = Move To
212 Front, as each value is used it's moved to the start of the
214 nSelectors = get_bits(bd, 15);
216 return RETVAL_DATA_ERROR;
217 for (i = 0; i < groupCount; i++)
219 for (i = 0; i < nSelectors; i++) {
221 for (j = 0; get_bits(bd, 1); j++)
223 return RETVAL_DATA_ERROR;
224 /* Decode MTF to get the next selector */
227 mtfSymbol[j] = mtfSymbol[j-1];
228 mtfSymbol[0] = selectors[i] = uc;
230 /* Read the Huffman coding tables for each group, which code
231 for symTotal literal symbols, plus two run symbols (RUNA,
233 symCount = symTotal+2;
234 for (j = 0; j < groupCount; j++) {
235 unsigned char length[MAX_SYMBOLS], temp[MAX_HUFCODE_BITS+1];
236 int minLen, maxLen, pp;
237 /* Read Huffman code lengths for each symbol. They're
238 stored in a way similar to mtf; record a starting
239 value for the first symbol, and an offset from the
240 previous value for everys symbol after that.
241 (Subtracting 1 before the loop and then adding it
242 back at the end is an optimization that makes the
243 test inside the loop simpler: symbol length 0
244 becomes negative, so an unsigned inequality catches
246 t = get_bits(bd, 5)-1;
247 for (i = 0; i < symCount; i++) {
249 if (((unsigned)t) > (MAX_HUFCODE_BITS-1))
250 return RETVAL_DATA_ERROR;
252 /* If first bit is 0, stop. Else
253 second bit indicates whether to
254 increment or decrement the value.
255 Optimization: grab 2 bits and unget
256 the second if the first was 0. */
263 /* Add one if second bit 1, else
264 * subtract 1. Avoids if/else */
267 /* Correct for the initial -1, to get the
268 * final symbol length */
271 /* Find largest and smallest lengths in this group */
272 minLen = maxLen = length[0];
274 for (i = 1; i < symCount; i++) {
275 if (length[i] > maxLen)
277 else if (length[i] < minLen)
281 /* Calculate permute[], base[], and limit[] tables from
284 * permute[] is the lookup table for converting
285 * Huffman coded symbols into decoded symbols. base[]
286 * is the amount to subtract from the value of a
287 * Huffman symbol of a given length when using
290 * limit[] indicates the largest numerical value a
291 * symbol with a given number of bits can have. This
292 * is how the Huffman codes can vary in length: each
293 * code with a value > limit[length] needs another
296 hufGroup = bd->groups+j;
297 hufGroup->minLen = minLen;
298 hufGroup->maxLen = maxLen;
299 /* Note that minLen can't be smaller than 1, so we
300 adjust the base and limit array pointers so we're
301 not always wasting the first entry. We do this
302 again when using them (during symbol decoding).*/
303 base = hufGroup->base-1;
304 limit = hufGroup->limit-1;
305 /* Calculate permute[]. Concurrently, initialize
306 * temp[] and limit[]. */
308 for (i = minLen; i <= maxLen; i++) {
309 temp[i] = limit[i] = 0;
310 for (t = 0; t < symCount; t++)
312 hufGroup->permute[pp++] = t;
314 /* Count symbols coded for at each bit length */
315 for (i = 0; i < symCount; i++)
317 /* Calculate limit[] (the largest symbol-coding value
318 *at each bit length, which is (previous limit <<
319 *1)+symbols at this level), and base[] (number of
320 *symbols to ignore at each bit length, which is limit
321 *minus the cumulative count of symbols coded for
324 for (i = minLen; i < maxLen; i++) {
326 /* We read the largest possible symbol size
327 and then unget bits after determining how
328 many we need, and those extra bits could be
329 set to anything. (They're noise from
330 future symbols.) At each level we're
331 really only interested in the first few
332 bits, so here we set all the trailing
333 to-be-ignored bits to 1 so they don't
334 affect the value > limit[length]
336 limit[i] = (pp << (maxLen - i)) - 1;
338 base[i+1] = pp-(t += temp[i]);
340 limit[maxLen+1] = INT_MAX; /* Sentinal value for
341 * reading next sym. */
342 limit[maxLen] = pp+temp[maxLen]-1;
345 /* We've finished reading and digesting the block header. Now
346 read this block's Huffman coded symbols from the file and
347 undo the Huffman coding and run length encoding, saving the
348 result into dbuf[dbufCount++] = uc */
350 /* Initialize symbol occurrence counters and symbol Move To
352 for (i = 0; i < 256; i++) {
354 mtfSymbol[i] = (unsigned char)i;
356 /* Loop through compressed symbols. */
357 runPos = dbufCount = symCount = selector = 0;
359 /* Determine which Huffman coding group to use. */
361 symCount = GROUP_SIZE-1;
362 if (selector >= nSelectors)
363 return RETVAL_DATA_ERROR;
364 hufGroup = bd->groups+selectors[selector++];
365 base = hufGroup->base-1;
366 limit = hufGroup->limit-1;
368 /* Read next Huffman-coded symbol. */
369 /* Note: It is far cheaper to read maxLen bits and
370 back up than it is to read minLen bits and then an
371 additional bit at a time, testing as we go.
372 Because there is a trailing last block (with file
373 CRC), there is no danger of the overread causing an
374 unexpected EOF for a valid compressed file. As a
375 further optimization, we do the read inline
376 (falling back to a call to get_bits if the buffer
377 runs dry). The following (up to got_huff_bits:) is
378 equivalent to j = get_bits(bd, hufGroup->maxLen);
380 while (bd->inbufBitCount < hufGroup->maxLen) {
381 if (bd->inbufPos == bd->inbufCount) {
382 j = get_bits(bd, hufGroup->maxLen);
386 (bd->inbufBits << 8)|bd->inbuf[bd->inbufPos++];
387 bd->inbufBitCount += 8;
389 bd->inbufBitCount -= hufGroup->maxLen;
390 j = (bd->inbufBits >> bd->inbufBitCount)&
391 ((1 << hufGroup->maxLen)-1);
393 /* Figure how many bits are in next symbol and
395 i = hufGroup->minLen;
398 bd->inbufBitCount += (hufGroup->maxLen - i);
399 /* Huffman decode value to get nextSym (with bounds checking) */
400 if ((i > hufGroup->maxLen)
401 || (((unsigned)(j = (j>>(hufGroup->maxLen-i))-base[i]))
403 return RETVAL_DATA_ERROR;
404 nextSym = hufGroup->permute[j];
405 /* We have now decoded the symbol, which indicates
406 either a new literal byte, or a repeated run of the
407 most recent literal byte. First, check if nextSym
408 indicates a repeated run, and if so loop collecting
409 how many times to repeat the last literal. */
410 if (((unsigned)nextSym) <= SYMBOL_RUNB) { /* RUNA or RUNB */
411 /* If this is the start of a new run, zero out
417 /* Neat trick that saves 1 symbol: instead of
418 or-ing 0 or 1 at each bit position, add 1
419 or 2 instead. For example, 1011 is 1 << 0
420 + 1 << 1 + 2 << 2. 1010 is 2 << 0 + 2 << 1
421 + 1 << 2. You can make any bit pattern
422 that way using 1 less symbol than the basic
423 or 0/1 method (except all bits 0, which
424 would use no symbols, but a run of length 0
425 doesn't mean anything in this context).
426 Thus space is saved. */
427 t += (runPos << nextSym);
428 /* +runPos if RUNA; +2*runPos if RUNB */
433 /* When we hit the first non-run symbol after a run,
434 we now know how many times to repeat the last
435 literal, so append that many copies to our buffer
436 of decoded symbols (dbuf) now. (The last literal
437 used is the one at the head of the mtfSymbol
441 if (dbufCount+t >= dbufSize)
442 return RETVAL_DATA_ERROR;
444 uc = symToByte[mtfSymbol[0]];
447 dbuf[dbufCount++] = uc;
449 /* Is this the terminating symbol? */
450 if (nextSym > symTotal)
452 /* At this point, nextSym indicates a new literal
453 character. Subtract one to get the position in the
454 MTF array at which this literal is currently to be
455 found. (Note that the result can't be -1 or 0,
456 because 0 and 1 are RUNA and RUNB. But another
457 instance of the first symbol in the mtf array,
458 position 0, would have been handled as part of a
459 run above. Therefore 1 unused mtf position minus 2
460 non-literal nextSym values equals -1.) */
461 if (dbufCount >= dbufSize)
462 return RETVAL_DATA_ERROR;
465 /* Adjust the MTF array. Since we typically expect to
466 *move only a small number of symbols, and are bound
467 *by 256 in any case, using memmove here would
468 *typically be bigger and slower due to function call
469 *overhead and other assorted setup costs. */
471 mtfSymbol[i] = mtfSymbol[i-1];
475 /* We have our literal byte. Save it into dbuf. */
477 dbuf[dbufCount++] = (unsigned int)uc;
479 /* At this point, we've read all the Huffman-coded symbols
480 (and repeated runs) for this block from the input stream,
481 and decoded them into the intermediate buffer. There are
482 dbufCount many decoded bytes in dbuf[]. Now undo the
483 Burrows-Wheeler transform on dbuf. See
484 http://dogma.net/markn/articles/bwt/bwt.htm
486 /* Turn byteCount into cumulative occurrence counts of 0 to n-1. */
488 for (i = 0; i < 256; i++) {
493 /* Figure out what order dbuf would be in if we sorted it. */
494 for (i = 0; i < dbufCount; i++) {
495 uc = (unsigned char)(dbuf[i] & 0xff);
496 dbuf[byteCount[uc]] |= (i << 8);
499 /* Decode first byte by hand to initialize "previous" byte.
500 Note that it doesn't get output, and if the first three
501 characters are identical it doesn't qualify as a run (hence
502 writeRunCountdown = 5). */
504 if (origPtr >= dbufCount)
505 return RETVAL_DATA_ERROR;
506 bd->writePos = dbuf[origPtr];
507 bd->writeCurrent = (unsigned char)(bd->writePos&0xff);
509 bd->writeRunCountdown = 5;
511 bd->writeCount = dbufCount;
516 /* Undo burrows-wheeler transform on intermediate buffer to produce output.
517 If start_bunzip was initialized with out_fd =-1, then up to len bytes of
518 data are written to outbuf. Return value is number of bytes written or
519 error (all errors are negative numbers). If out_fd!=-1, outbuf and len
520 are ignored, data is written to out_fd and return is RETVAL_OK or error.
523 static int INIT read_bunzip(struct bunzip_data *bd, char *outbuf, int len)
525 const unsigned int *dbuf;
526 int pos, xcurrent, previous, gotcount;
528 /* If last read was short due to end of file, return last block now */
529 if (bd->writeCount < 0)
530 return bd->writeCount;
535 xcurrent = bd->writeCurrent;
537 /* We will always have pending decoded data to write into the output
538 buffer unless this is the very first call (in which case we haven't
539 Huffman-decoded a block into the intermediate buffer yet). */
541 if (bd->writeCopies) {
542 /* Inside the loop, writeCopies means extra copies (beyond 1) */
544 /* Loop outputting bytes */
546 /* If the output buffer is full, snapshot
547 * state and return */
548 if (gotcount >= len) {
550 bd->writeCurrent = xcurrent;
554 /* Write next byte into output buffer, updating CRC */
555 outbuf[gotcount++] = xcurrent;
556 bd->writeCRC = (((bd->writeCRC) << 8)
557 ^bd->crc32Table[((bd->writeCRC) >> 24)
559 /* Loop now if we're outputting multiple
560 * copies of this byte */
561 if (bd->writeCopies) {
566 if (!bd->writeCount--)
568 /* Follow sequence vector to undo
569 * Burrows-Wheeler transform */
574 /* After 3 consecutive copies of the same
575 byte, the 4th is a repeat count. We count
576 down from 4 instead *of counting up because
577 testing for non-zero is faster */
578 if (--bd->writeRunCountdown) {
579 if (xcurrent != previous)
580 bd->writeRunCountdown = 4;
582 /* We have a repeated run, this byte
583 * indicates the count */
584 bd->writeCopies = xcurrent;
586 bd->writeRunCountdown = 5;
587 /* Sometimes there are just 3 bytes
589 if (!bd->writeCopies)
590 goto decode_next_byte;
591 /* Subtract the 1 copy we'd output
592 * anyway to get extras */
596 /* Decompression of this block completed successfully */
597 bd->writeCRC = ~bd->writeCRC;
598 bd->totalCRC = ((bd->totalCRC << 1) |
599 (bd->totalCRC >> 31)) ^ bd->writeCRC;
600 /* If this block had a CRC error, force file level CRC error. */
601 if (bd->writeCRC != bd->headerCRC) {
602 bd->totalCRC = bd->headerCRC+1;
603 return RETVAL_LAST_BLOCK;
607 /* Refill the intermediate buffer by Huffman-decoding next
609 /* (previous is just a convenient unused temp variable here) */
610 previous = get_next_block(bd);
612 bd->writeCount = previous;
613 return (previous != RETVAL_LAST_BLOCK) ? previous : gotcount;
615 bd->writeCRC = 0xffffffffUL;
617 xcurrent = bd->writeCurrent;
618 goto decode_next_byte;
621 static long INIT nofill(void *buf, unsigned long len)
626 /* Allocate the structure, read file header. If in_fd ==-1, inbuf must contain
627 a complete bunzip file (len bytes long). If in_fd!=-1, inbuf and len are
628 ignored, and data is read from file handle into temporary buffer. */
629 static int INIT start_bunzip(struct bunzip_data **bdp, void *inbuf, long len,
630 long (*fill)(void*, unsigned long))
632 struct bunzip_data *bd;
633 unsigned int i, j, c;
634 const unsigned int BZh0 =
635 (((unsigned int)'B') << 24)+(((unsigned int)'Z') << 16)
636 +(((unsigned int)'h') << 8)+(unsigned int)'0';
638 /* Figure out how much data to allocate */
639 i = sizeof(struct bunzip_data);
641 /* Allocate bunzip_data. Most fields initialize to zero. */
642 bd = *bdp = malloc(i);
644 return RETVAL_OUT_OF_MEMORY;
645 memset(bd, 0, sizeof(struct bunzip_data));
646 /* Setup input buffer */
648 bd->inbufCount = len;
654 /* Init the CRC32 table (big endian) */
655 for (i = 0; i < 256; i++) {
658 c = c&0x80000000 ? (c << 1)^(CRC32_POLY_BE) : (c << 1);
659 bd->crc32Table[i] = c;
662 /* Ensure that file starts with "BZh['1'-'9']." */
663 i = get_bits(bd, 32);
664 if (((unsigned int)(i-BZh0-1)) >= 9)
665 return RETVAL_NOT_BZIP_DATA;
667 /* Fourth byte (ascii '1'-'9'), indicates block size in units of 100k of
668 uncompressed data. Allocate intermediate buffer for block. */
669 bd->dbufSize = 100000*(i-BZh0);
671 bd->dbuf = large_malloc(bd->dbufSize * sizeof(int));
673 return RETVAL_OUT_OF_MEMORY;
677 /* Example usage: decompress src_fd to dst_fd. (Stops at end of bzip2 data,
679 STATIC int INIT bunzip2(unsigned char *buf, long len,
680 long (*fill)(void*, unsigned long),
681 long (*flush)(void*, unsigned long),
682 unsigned char *outbuf,
684 void(*error)(char *x))
686 struct bunzip_data *bd;
688 unsigned char *inbuf;
691 outbuf = malloc(BZIP2_IOBUF_SIZE);
694 error("Could not allocate output buffer");
695 return RETVAL_OUT_OF_MEMORY;
700 inbuf = malloc(BZIP2_IOBUF_SIZE);
702 error("Could not allocate input buffer");
703 i = RETVAL_OUT_OF_MEMORY;
706 i = start_bunzip(&bd, inbuf, len, fill);
709 i = read_bunzip(bd, outbuf, BZIP2_IOBUF_SIZE);
715 if (i != flush(outbuf, i)) {
716 i = RETVAL_UNEXPECTED_OUTPUT_EOF;
721 /* Check CRC and release memory */
722 if (i == RETVAL_LAST_BLOCK) {
723 if (bd->headerCRC != bd->totalCRC)
724 error("Data integrity error when decompressing.");
727 } else if (i == RETVAL_UNEXPECTED_OUTPUT_EOF) {
728 error("Compressed file ends unexpectedly");
733 large_free(bd->dbuf);
747 STATIC int INIT __decompress(unsigned char *buf, long len,
748 long (*fill)(void*, unsigned long),
749 long (*flush)(void*, unsigned long),
750 unsigned char *outbuf, long olen,
752 void (*error)(char *x))
754 return bunzip2(buf, len - 4, fill, flush, outbuf, pos, error);