1 /* vi: set sw = 4 ts = 4: */
2 /* Small bzip2 deflate implementation, by Rob Landley (rob@landley.net).
4 Based on bzip2 decompression code by Julian R Seward (jseward@acm.org),
5 which also acknowledges contributions by Mike Burrows, David Wheeler,
6 Peter Fenwick, Alistair Moffat, Radford Neal, Ian H. Witten,
7 Robert Sedgewick, and Jon L. Bentley.
9 This code is licensed under the LGPLv2:
10 LGPL (http://www.gnu.org/copyleft/lgpl.html
14 Size and speed optimizations by Manuel Novoa III (mjn3@codepoet.org).
16 More efficient reading of Huffman codes, a streamlined read_bunzip()
17 function, and various other tweaks. In (limited) tests, approximately
18 20% faster than bzcat on x86 and about 10% faster on arm.
20 Note that about 2/3 of the time is spent in read_unzip() reversing
21 the Burrows-Wheeler transformation. Much of that time is delay
22 resulting from cache misses.
24 I would ask that anyone benefiting from this work, especially those
25 using it in commercial products, consider making a donation to my local
26 non-profit hospice organization in the name of the woman I loved, who
27 passed away Feb. 12, 2003.
29 In memory of Toni W. Hagan
31 Hospice of Acadiana, Inc.
32 2600 Johnston St., Suite 200
33 Lafayette, LA 70503-3240
35 Phone (337) 232-1234 or 1-800-738-2226
38 http://www.hospiceacadiana.com/
44 Made it fit for running in Linux Kernel by Alain Knaff (alain@knaff.lu)
51 #include <linux/decompress/bunzip2.h>
52 #include <linux/slab.h>
55 #include <linux/decompress/mm.h>
58 #define INT_MAX 0x7fffffff
61 /* Constants for Huffman coding */
63 #define GROUP_SIZE 50 /* 64 would have been more efficient */
64 #define MAX_HUFCODE_BITS 20 /* Longest Huffman code allowed */
65 #define MAX_SYMBOLS 258 /* 256 literals + RUNA + RUNB */
69 /* Status return values */
71 #define RETVAL_LAST_BLOCK (-1)
72 #define RETVAL_NOT_BZIP_DATA (-2)
73 #define RETVAL_UNEXPECTED_INPUT_EOF (-3)
74 #define RETVAL_UNEXPECTED_OUTPUT_EOF (-4)
75 #define RETVAL_DATA_ERROR (-5)
76 #define RETVAL_OUT_OF_MEMORY (-6)
77 #define RETVAL_OBSOLETE_INPUT (-7)
79 /* Other housekeeping constants */
80 #define BZIP2_IOBUF_SIZE 4096
82 /* This is what we know about each Huffman coding group */
84 /* We have an extra slot at the end of limit[] for a sentinal value. */
85 int limit[MAX_HUFCODE_BITS+1];
86 int base[MAX_HUFCODE_BITS];
87 int permute[MAX_SYMBOLS];
91 /* Structure holding all the housekeeping data, including IO buffers and
92 memory that persists between calls to bunzip */
94 /* State for interrupting output loop */
95 int writeCopies, writePos, writeRunCountdown, writeCount, writeCurrent;
96 /* I/O tracking data (file handles, buffers, positions, etc.) */
97 int (*fill)(void*, unsigned int);
98 int inbufCount, inbufPos /*, outbufPos*/;
99 unsigned char *inbuf /*,*outbuf*/;
100 unsigned int inbufBitCount, inbufBits;
101 /* The CRC values stored in the block header and calculated from the
103 unsigned int crc32Table[256], headerCRC, totalCRC, writeCRC;
104 /* Intermediate buffer and its size (in bytes) */
105 unsigned int *dbuf, dbufSize;
106 /* These things are a bit too big to go on the stack */
107 unsigned char selectors[32768]; /* nSelectors = 15 bits */
108 struct group_data groups[MAX_GROUPS]; /* Huffman coding tables */
109 int io_error; /* non-zero if we have IO error */
113 /* Return the next nnn bits of input. All reads from the compressed input
114 are done through this function. All reads are big endian */
115 static unsigned int INIT get_bits(struct bunzip_data *bd, char bits_wanted)
117 unsigned int bits = 0;
119 /* If we need to get more data from the byte buffer, do so.
120 (Loop getting one byte at a time to enforce endianness and avoid
121 unaligned access.) */
122 while (bd->inbufBitCount < bits_wanted) {
123 /* If we need to read more data from file into byte buffer, do
125 if (bd->inbufPos == bd->inbufCount) {
128 bd->inbufCount = bd->fill(bd->inbuf, BZIP2_IOBUF_SIZE);
129 if (bd->inbufCount <= 0) {
130 bd->io_error = RETVAL_UNEXPECTED_INPUT_EOF;
135 /* Avoid 32-bit overflow (dump bit buffer to top of output) */
136 if (bd->inbufBitCount >= 24) {
137 bits = bd->inbufBits&((1 << bd->inbufBitCount)-1);
138 bits_wanted -= bd->inbufBitCount;
139 bits <<= bits_wanted;
140 bd->inbufBitCount = 0;
142 /* Grab next 8 bits of input from buffer. */
143 bd->inbufBits = (bd->inbufBits << 8)|bd->inbuf[bd->inbufPos++];
144 bd->inbufBitCount += 8;
146 /* Calculate result */
147 bd->inbufBitCount -= bits_wanted;
148 bits |= (bd->inbufBits >> bd->inbufBitCount)&((1 << bits_wanted)-1);
153 /* Unpacks the next block and sets up for the inverse burrows-wheeler step. */
155 static int INIT get_next_block(struct bunzip_data *bd)
157 struct group_data *hufGroup = NULL;
160 int dbufCount, nextSym, dbufSize, groupCount, selector,
161 i, j, k, t, runPos, symCount, symTotal, nSelectors,
163 unsigned char uc, symToByte[256], mtfSymbol[256], *selectors;
164 unsigned int *dbuf, origPtr;
167 dbufSize = bd->dbufSize;
168 selectors = bd->selectors;
170 /* Read in header signature and CRC, then validate signature.
171 (last block signature means CRC is for whole file, return now) */
172 i = get_bits(bd, 24);
173 j = get_bits(bd, 24);
174 bd->headerCRC = get_bits(bd, 32);
175 if ((i == 0x177245) && (j == 0x385090))
176 return RETVAL_LAST_BLOCK;
177 if ((i != 0x314159) || (j != 0x265359))
178 return RETVAL_NOT_BZIP_DATA;
179 /* We can add support for blockRandomised if anybody complains.
180 There was some code for this in busybox 1.0.0-pre3, but nobody ever
181 noticed that it didn't actually work. */
183 return RETVAL_OBSOLETE_INPUT;
184 origPtr = get_bits(bd, 24);
185 if (origPtr > dbufSize)
186 return RETVAL_DATA_ERROR;
187 /* mapping table: if some byte values are never used (encoding things
188 like ascii text), the compression code removes the gaps to have fewer
189 symbols to deal with, and writes a sparse bitfield indicating which
190 values were present. We make a translation table to convert the
191 symbols back to the corresponding bytes. */
192 t = get_bits(bd, 16);
194 for (i = 0; i < 16; i++) {
195 if (t&(1 << (15-i))) {
196 k = get_bits(bd, 16);
197 for (j = 0; j < 16; j++)
199 symToByte[symTotal++] = (16*i)+j;
202 /* How many different Huffman coding groups does this block use? */
203 groupCount = get_bits(bd, 3);
204 if (groupCount < 2 || groupCount > MAX_GROUPS)
205 return RETVAL_DATA_ERROR;
206 /* nSelectors: Every GROUP_SIZE many symbols we select a new
207 Huffman coding group. Read in the group selector list,
208 which is stored as MTF encoded bit runs. (MTF = Move To
209 Front, as each value is used it's moved to the start of the
211 nSelectors = get_bits(bd, 15);
213 return RETVAL_DATA_ERROR;
214 for (i = 0; i < groupCount; i++)
216 for (i = 0; i < nSelectors; i++) {
218 for (j = 0; get_bits(bd, 1); j++)
220 return RETVAL_DATA_ERROR;
221 /* Decode MTF to get the next selector */
224 mtfSymbol[j] = mtfSymbol[j-1];
225 mtfSymbol[0] = selectors[i] = uc;
227 /* Read the Huffman coding tables for each group, which code
228 for symTotal literal symbols, plus two run symbols (RUNA,
230 symCount = symTotal+2;
231 for (j = 0; j < groupCount; j++) {
232 unsigned char length[MAX_SYMBOLS], temp[MAX_HUFCODE_BITS+1];
233 int minLen, maxLen, pp;
234 /* Read Huffman code lengths for each symbol. They're
235 stored in a way similar to mtf; record a starting
236 value for the first symbol, and an offset from the
237 previous value for everys symbol after that.
238 (Subtracting 1 before the loop and then adding it
239 back at the end is an optimization that makes the
240 test inside the loop simpler: symbol length 0
241 becomes negative, so an unsigned inequality catches
243 t = get_bits(bd, 5)-1;
244 for (i = 0; i < symCount; i++) {
246 if (((unsigned)t) > (MAX_HUFCODE_BITS-1))
247 return RETVAL_DATA_ERROR;
249 /* If first bit is 0, stop. Else
250 second bit indicates whether to
251 increment or decrement the value.
252 Optimization: grab 2 bits and unget
253 the second if the first was 0. */
260 /* Add one if second bit 1, else
261 * subtract 1. Avoids if/else */
264 /* Correct for the initial -1, to get the
265 * final symbol length */
268 /* Find largest and smallest lengths in this group */
269 minLen = maxLen = length[0];
271 for (i = 1; i < symCount; i++) {
272 if (length[i] > maxLen)
274 else if (length[i] < minLen)
278 /* Calculate permute[], base[], and limit[] tables from
281 * permute[] is the lookup table for converting
282 * Huffman coded symbols into decoded symbols. base[]
283 * is the amount to subtract from the value of a
284 * Huffman symbol of a given length when using
287 * limit[] indicates the largest numerical value a
288 * symbol with a given number of bits can have. This
289 * is how the Huffman codes can vary in length: each
290 * code with a value > limit[length] needs another
293 hufGroup = bd->groups+j;
294 hufGroup->minLen = minLen;
295 hufGroup->maxLen = maxLen;
296 /* Note that minLen can't be smaller than 1, so we
297 adjust the base and limit array pointers so we're
298 not always wasting the first entry. We do this
299 again when using them (during symbol decoding).*/
300 base = hufGroup->base-1;
301 limit = hufGroup->limit-1;
302 /* Calculate permute[]. Concurrently, initialize
303 * temp[] and limit[]. */
305 for (i = minLen; i <= maxLen; i++) {
306 temp[i] = limit[i] = 0;
307 for (t = 0; t < symCount; t++)
309 hufGroup->permute[pp++] = t;
311 /* Count symbols coded for at each bit length */
312 for (i = 0; i < symCount; i++)
314 /* Calculate limit[] (the largest symbol-coding value
315 *at each bit length, which is (previous limit <<
316 *1)+symbols at this level), and base[] (number of
317 *symbols to ignore at each bit length, which is limit
318 *minus the cumulative count of symbols coded for
321 for (i = minLen; i < maxLen; i++) {
323 /* We read the largest possible symbol size
324 and then unget bits after determining how
325 many we need, and those extra bits could be
326 set to anything. (They're noise from
327 future symbols.) At each level we're
328 really only interested in the first few
329 bits, so here we set all the trailing
330 to-be-ignored bits to 1 so they don't
331 affect the value > limit[length]
333 limit[i] = (pp << (maxLen - i)) - 1;
335 base[i+1] = pp-(t += temp[i]);
337 limit[maxLen+1] = INT_MAX; /* Sentinal value for
338 * reading next sym. */
339 limit[maxLen] = pp+temp[maxLen]-1;
342 /* We've finished reading and digesting the block header. Now
343 read this block's Huffman coded symbols from the file and
344 undo the Huffman coding and run length encoding, saving the
345 result into dbuf[dbufCount++] = uc */
347 /* Initialize symbol occurrence counters and symbol Move To
349 for (i = 0; i < 256; i++) {
351 mtfSymbol[i] = (unsigned char)i;
353 /* Loop through compressed symbols. */
354 runPos = dbufCount = symCount = selector = 0;
356 /* Determine which Huffman coding group to use. */
358 symCount = GROUP_SIZE-1;
359 if (selector >= nSelectors)
360 return RETVAL_DATA_ERROR;
361 hufGroup = bd->groups+selectors[selector++];
362 base = hufGroup->base-1;
363 limit = hufGroup->limit-1;
365 /* Read next Huffman-coded symbol. */
366 /* Note: It is far cheaper to read maxLen bits and
367 back up than it is to read minLen bits and then an
368 additional bit at a time, testing as we go.
369 Because there is a trailing last block (with file
370 CRC), there is no danger of the overread causing an
371 unexpected EOF for a valid compressed file. As a
372 further optimization, we do the read inline
373 (falling back to a call to get_bits if the buffer
374 runs dry). The following (up to got_huff_bits:) is
375 equivalent to j = get_bits(bd, hufGroup->maxLen);
377 while (bd->inbufBitCount < hufGroup->maxLen) {
378 if (bd->inbufPos == bd->inbufCount) {
379 j = get_bits(bd, hufGroup->maxLen);
383 (bd->inbufBits << 8)|bd->inbuf[bd->inbufPos++];
384 bd->inbufBitCount += 8;
386 bd->inbufBitCount -= hufGroup->maxLen;
387 j = (bd->inbufBits >> bd->inbufBitCount)&
388 ((1 << hufGroup->maxLen)-1);
390 /* Figure how how many bits are in next symbol and
392 i = hufGroup->minLen;
395 bd->inbufBitCount += (hufGroup->maxLen - i);
396 /* Huffman decode value to get nextSym (with bounds checking) */
397 if ((i > hufGroup->maxLen)
398 || (((unsigned)(j = (j>>(hufGroup->maxLen-i))-base[i]))
400 return RETVAL_DATA_ERROR;
401 nextSym = hufGroup->permute[j];
402 /* We have now decoded the symbol, which indicates
403 either a new literal byte, or a repeated run of the
404 most recent literal byte. First, check if nextSym
405 indicates a repeated run, and if so loop collecting
406 how many times to repeat the last literal. */
407 if (((unsigned)nextSym) <= SYMBOL_RUNB) { /* RUNA or RUNB */
408 /* If this is the start of a new run, zero out
414 /* Neat trick that saves 1 symbol: instead of
415 or-ing 0 or 1 at each bit position, add 1
416 or 2 instead. For example, 1011 is 1 << 0
417 + 1 << 1 + 2 << 2. 1010 is 2 << 0 + 2 << 1
418 + 1 << 2. You can make any bit pattern
419 that way using 1 less symbol than the basic
420 or 0/1 method (except all bits 0, which
421 would use no symbols, but a run of length 0
422 doesn't mean anything in this context).
423 Thus space is saved. */
424 t += (runPos << nextSym);
425 /* +runPos if RUNA; +2*runPos if RUNB */
430 /* When we hit the first non-run symbol after a run,
431 we now know how many times to repeat the last
432 literal, so append that many copies to our buffer
433 of decoded symbols (dbuf) now. (The last literal
434 used is the one at the head of the mtfSymbol
438 if (dbufCount+t >= dbufSize)
439 return RETVAL_DATA_ERROR;
441 uc = symToByte[mtfSymbol[0]];
444 dbuf[dbufCount++] = uc;
446 /* Is this the terminating symbol? */
447 if (nextSym > symTotal)
449 /* At this point, nextSym indicates a new literal
450 character. Subtract one to get the position in the
451 MTF array at which this literal is currently to be
452 found. (Note that the result can't be -1 or 0,
453 because 0 and 1 are RUNA and RUNB. But another
454 instance of the first symbol in the mtf array,
455 position 0, would have been handled as part of a
456 run above. Therefore 1 unused mtf position minus 2
457 non-literal nextSym values equals -1.) */
458 if (dbufCount >= dbufSize)
459 return RETVAL_DATA_ERROR;
462 /* Adjust the MTF array. Since we typically expect to
463 *move only a small number of symbols, and are bound
464 *by 256 in any case, using memmove here would
465 *typically be bigger and slower due to function call
466 *overhead and other assorted setup costs. */
468 mtfSymbol[i] = mtfSymbol[i-1];
472 /* We have our literal byte. Save it into dbuf. */
474 dbuf[dbufCount++] = (unsigned int)uc;
476 /* At this point, we've read all the Huffman-coded symbols
477 (and repeated runs) for this block from the input stream,
478 and decoded them into the intermediate buffer. There are
479 dbufCount many decoded bytes in dbuf[]. Now undo the
480 Burrows-Wheeler transform on dbuf. See
481 http://dogma.net/markn/articles/bwt/bwt.htm
483 /* Turn byteCount into cumulative occurrence counts of 0 to n-1. */
485 for (i = 0; i < 256; i++) {
490 /* Figure out what order dbuf would be in if we sorted it. */
491 for (i = 0; i < dbufCount; i++) {
492 uc = (unsigned char)(dbuf[i] & 0xff);
493 dbuf[byteCount[uc]] |= (i << 8);
496 /* Decode first byte by hand to initialize "previous" byte.
497 Note that it doesn't get output, and if the first three
498 characters are identical it doesn't qualify as a run (hence
499 writeRunCountdown = 5). */
501 if (origPtr >= dbufCount)
502 return RETVAL_DATA_ERROR;
503 bd->writePos = dbuf[origPtr];
504 bd->writeCurrent = (unsigned char)(bd->writePos&0xff);
506 bd->writeRunCountdown = 5;
508 bd->writeCount = dbufCount;
513 /* Undo burrows-wheeler transform on intermediate buffer to produce output.
514 If start_bunzip was initialized with out_fd =-1, then up to len bytes of
515 data are written to outbuf. Return value is number of bytes written or
516 error (all errors are negative numbers). If out_fd!=-1, outbuf and len
517 are ignored, data is written to out_fd and return is RETVAL_OK or error.
520 static int INIT read_bunzip(struct bunzip_data *bd, char *outbuf, int len)
522 const unsigned int *dbuf;
523 int pos, xcurrent, previous, gotcount;
525 /* If last read was short due to end of file, return last block now */
526 if (bd->writeCount < 0)
527 return bd->writeCount;
532 xcurrent = bd->writeCurrent;
534 /* We will always have pending decoded data to write into the output
535 buffer unless this is the very first call (in which case we haven't
536 Huffman-decoded a block into the intermediate buffer yet). */
538 if (bd->writeCopies) {
539 /* Inside the loop, writeCopies means extra copies (beyond 1) */
541 /* Loop outputting bytes */
543 /* If the output buffer is full, snapshot
544 * state and return */
545 if (gotcount >= len) {
547 bd->writeCurrent = xcurrent;
551 /* Write next byte into output buffer, updating CRC */
552 outbuf[gotcount++] = xcurrent;
553 bd->writeCRC = (((bd->writeCRC) << 8)
554 ^bd->crc32Table[((bd->writeCRC) >> 24)
556 /* Loop now if we're outputting multiple
557 * copies of this byte */
558 if (bd->writeCopies) {
563 if (!bd->writeCount--)
565 /* Follow sequence vector to undo
566 * Burrows-Wheeler transform */
571 /* After 3 consecutive copies of the same
572 byte, the 4th is a repeat count. We count
573 down from 4 instead *of counting up because
574 testing for non-zero is faster */
575 if (--bd->writeRunCountdown) {
576 if (xcurrent != previous)
577 bd->writeRunCountdown = 4;
579 /* We have a repeated run, this byte
580 * indicates the count */
581 bd->writeCopies = xcurrent;
583 bd->writeRunCountdown = 5;
584 /* Sometimes there are just 3 bytes
586 if (!bd->writeCopies)
587 goto decode_next_byte;
588 /* Subtract the 1 copy we'd output
589 * anyway to get extras */
593 /* Decompression of this block completed successfully */
594 bd->writeCRC = ~bd->writeCRC;
595 bd->totalCRC = ((bd->totalCRC << 1) |
596 (bd->totalCRC >> 31)) ^ bd->writeCRC;
597 /* If this block had a CRC error, force file level CRC error. */
598 if (bd->writeCRC != bd->headerCRC) {
599 bd->totalCRC = bd->headerCRC+1;
600 return RETVAL_LAST_BLOCK;
604 /* Refill the intermediate buffer by Huffman-decoding next
606 /* (previous is just a convenient unused temp variable here) */
607 previous = get_next_block(bd);
609 bd->writeCount = previous;
610 return (previous != RETVAL_LAST_BLOCK) ? previous : gotcount;
612 bd->writeCRC = 0xffffffffUL;
614 xcurrent = bd->writeCurrent;
615 goto decode_next_byte;
618 static int INIT nofill(void *buf, unsigned int len)
623 /* Allocate the structure, read file header. If in_fd ==-1, inbuf must contain
624 a complete bunzip file (len bytes long). If in_fd!=-1, inbuf and len are
625 ignored, and data is read from file handle into temporary buffer. */
626 static int INIT start_bunzip(struct bunzip_data **bdp, void *inbuf, int len,
627 int (*fill)(void*, unsigned int))
629 struct bunzip_data *bd;
630 unsigned int i, j, c;
631 const unsigned int BZh0 =
632 (((unsigned int)'B') << 24)+(((unsigned int)'Z') << 16)
633 +(((unsigned int)'h') << 8)+(unsigned int)'0';
635 /* Figure out how much data to allocate */
636 i = sizeof(struct bunzip_data);
638 /* Allocate bunzip_data. Most fields initialize to zero. */
639 bd = *bdp = malloc(i);
640 memset(bd, 0, sizeof(struct bunzip_data));
641 /* Setup input buffer */
643 bd->inbufCount = len;
649 /* Init the CRC32 table (big endian) */
650 for (i = 0; i < 256; i++) {
653 c = c&0x80000000 ? (c << 1)^0x04c11db7 : (c << 1);
654 bd->crc32Table[i] = c;
657 /* Ensure that file starts with "BZh['1'-'9']." */
658 i = get_bits(bd, 32);
659 if (((unsigned int)(i-BZh0-1)) >= 9)
660 return RETVAL_NOT_BZIP_DATA;
662 /* Fourth byte (ascii '1'-'9'), indicates block size in units of 100k of
663 uncompressed data. Allocate intermediate buffer for block. */
664 bd->dbufSize = 100000*(i-BZh0);
666 bd->dbuf = large_malloc(bd->dbufSize * sizeof(int));
670 /* Example usage: decompress src_fd to dst_fd. (Stops at end of bzip2 data,
672 STATIC int INIT bunzip2(unsigned char *buf, int len,
673 int(*fill)(void*, unsigned int),
674 int(*flush)(void*, unsigned int),
675 unsigned char *outbuf,
677 void(*error_fn)(char *x))
679 struct bunzip_data *bd;
681 unsigned char *inbuf;
683 set_error_fn(error_fn);
685 outbuf = malloc(BZIP2_IOBUF_SIZE);
688 error("Could not allocate output bufer");
694 inbuf = malloc(BZIP2_IOBUF_SIZE);
696 error("Could not allocate input bufer");
699 i = start_bunzip(&bd, inbuf, len, fill);
702 i = read_bunzip(bd, outbuf, BZIP2_IOBUF_SIZE);
708 if (i != flush(outbuf, i)) {
709 i = RETVAL_UNEXPECTED_OUTPUT_EOF;
714 /* Check CRC and release memory */
715 if (i == RETVAL_LAST_BLOCK) {
716 if (bd->headerCRC != bd->totalCRC)
717 error("Data integrity error when decompressing.");
720 } else if (i == RETVAL_UNEXPECTED_OUTPUT_EOF) {
721 error("Compressed file ends unexpectedly");
724 large_free(bd->dbuf);
737 STATIC int INIT decompress(unsigned char *buf, int len,
738 int(*fill)(void*, unsigned int),
739 int(*flush)(void*, unsigned int),
740 unsigned char *outbuf,
742 void(*error_fn)(char *x))
744 return bunzip2(buf, len - 4, fill, flush, outbuf, pos, error_fn);