blob: fe5953da2412df5c428e281c3c50a86d735db735
1 | /* vi: set sw=4 ts=4: */ |
2 | /* Small bzip2 deflate implementation, by Rob Landley (rob@landley.net). |
3 | |
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. |
8 | |
9 | Licensed under GPLv2 or later, see file LICENSE in this source tree. |
10 | */ |
11 | |
12 | /* |
13 | Size and speed optimizations by Manuel Novoa III (mjn3@codepoet.org). |
14 | |
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. |
18 | |
19 | Note that about 2/3 of the time is spent in read_bunzip() reversing |
20 | the Burrows-Wheeler transformation. Much of that time is delay |
21 | resulting from cache misses. |
22 | |
23 | (2010 update by vda: profiled "bzcat <84mbyte.bz2 >/dev/null" |
24 | on x86-64 CPU with L2 > 1M: get_next_block is hotter than read_bunzip: |
25 | %time seconds calls function |
26 | 71.01 12.69 444 get_next_block |
27 | 28.65 5.12 93065 read_bunzip |
28 | 00.22 0.04 7736490 get_bits |
29 | 00.11 0.02 47 dealloc_bunzip |
30 | 00.00 0.00 93018 full_write |
31 | ...) |
32 | |
33 | |
34 | I would ask that anyone benefiting from this work, especially those |
35 | using it in commercial products, consider making a donation to my local |
36 | non-profit hospice organization (www.hospiceacadiana.com) in the name of |
37 | the woman I loved, Toni W. Hagan, who passed away Feb. 12, 2003. |
38 | |
39 | Manuel |
40 | */ |
41 | |
42 | #include "libbb.h" |
43 | #include "bb_archive.h" |
44 | |
45 | #if 0 |
46 | # define dbg(...) bb_error_msg(__VA_ARGS__) |
47 | #else |
48 | # define dbg(...) ((void)0) |
49 | #endif |
50 | |
51 | /* Constants for Huffman coding */ |
52 | #define MAX_GROUPS 6 |
53 | #define GROUP_SIZE 50 /* 64 would have been more efficient */ |
54 | #define MAX_HUFCODE_BITS 20 /* Longest Huffman code allowed */ |
55 | #define MAX_SYMBOLS 258 /* 256 literals + RUNA + RUNB */ |
56 | #define SYMBOL_RUNA 0 |
57 | #define SYMBOL_RUNB 1 |
58 | |
59 | /* Status return values */ |
60 | #define RETVAL_OK 0 |
61 | #define RETVAL_LAST_BLOCK (dbg("%d", __LINE__), -1) |
62 | #define RETVAL_NOT_BZIP_DATA (dbg("%d", __LINE__), -2) |
63 | #define RETVAL_UNEXPECTED_INPUT_EOF (dbg("%d", __LINE__), -3) |
64 | #define RETVAL_SHORT_WRITE (dbg("%d", __LINE__), -4) |
65 | #define RETVAL_DATA_ERROR (dbg("%d", __LINE__), -5) |
66 | #define RETVAL_OUT_OF_MEMORY (dbg("%d", __LINE__), -6) |
67 | #define RETVAL_OBSOLETE_INPUT (dbg("%d", __LINE__), -7) |
68 | |
69 | /* Other housekeeping constants */ |
70 | #define IOBUF_SIZE 4096 |
71 | |
72 | /* This is what we know about each Huffman coding group */ |
73 | struct group_data { |
74 | /* We have an extra slot at the end of limit[] for a sentinel value. */ |
75 | int limit[MAX_HUFCODE_BITS+1], base[MAX_HUFCODE_BITS], permute[MAX_SYMBOLS]; |
76 | int minLen, maxLen; |
77 | }; |
78 | |
79 | /* Structure holding all the housekeeping data, including IO buffers and |
80 | * memory that persists between calls to bunzip |
81 | * Found the most used member: |
82 | * cat this_file.c | sed -e 's/"/ /g' -e "s/'/ /g" | xargs -n1 \ |
83 | * | grep 'bd->' | sed 's/^.*bd->/bd->/' | sort | $PAGER |
84 | * and moved it (inbufBitCount) to offset 0. |
85 | */ |
86 | struct bunzip_data { |
87 | /* I/O tracking data (file handles, buffers, positions, etc.) */ |
88 | unsigned inbufBitCount, inbufBits; |
89 | int in_fd, out_fd, inbufCount, inbufPos /*, outbufPos*/; |
90 | uint8_t *inbuf /*,*outbuf*/; |
91 | |
92 | /* State for interrupting output loop */ |
93 | int writeCopies, writePos, writeRunCountdown, writeCount; |
94 | int writeCurrent; /* actually a uint8_t */ |
95 | |
96 | /* The CRC values stored in the block header and calculated from the data */ |
97 | uint32_t headerCRC, totalCRC, writeCRC; |
98 | |
99 | /* Intermediate buffer and its size (in bytes) */ |
100 | uint32_t *dbuf; |
101 | unsigned dbufSize; |
102 | |
103 | /* For I/O error handling */ |
104 | jmp_buf jmpbuf; |
105 | |
106 | /* Big things go last (register-relative addressing can be larger for big offsets) */ |
107 | uint32_t crc32Table[256]; |
108 | uint8_t selectors[32768]; /* nSelectors=15 bits */ |
109 | struct group_data groups[MAX_GROUPS]; /* Huffman coding tables */ |
110 | }; |
111 | /* typedef struct bunzip_data bunzip_data; -- done in .h file */ |
112 | |
113 | |
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 get_bits(bunzip_data *bd, int bits_wanted) |
117 | { |
118 | unsigned bits = 0; |
119 | /* Cache bd->inbufBitCount in a CPU register (hopefully): */ |
120 | int bit_count = bd->inbufBitCount; |
121 | |
122 | /* If we need to get more data from the byte buffer, do so. (Loop getting |
123 | one byte at a time to enforce endianness and avoid unaligned access.) */ |
124 | while (bit_count < bits_wanted) { |
125 | |
126 | /* If we need to read more data from file into byte buffer, do so */ |
127 | if (bd->inbufPos == bd->inbufCount) { |
128 | /* if "no input fd" case: in_fd == -1, read fails, we jump */ |
129 | bd->inbufCount = read(bd->in_fd, bd->inbuf, IOBUF_SIZE); |
130 | if (bd->inbufCount <= 0) |
131 | longjmp(bd->jmpbuf, RETVAL_UNEXPECTED_INPUT_EOF); |
132 | bd->inbufPos = 0; |
133 | } |
134 | |
135 | /* Avoid 32-bit overflow (dump bit buffer to top of output) */ |
136 | if (bit_count >= 24) { |
137 | bits = bd->inbufBits & ((1 << bit_count) - 1); |
138 | bits_wanted -= bit_count; |
139 | bits <<= bits_wanted; |
140 | bit_count = 0; |
141 | } |
142 | |
143 | /* Grab next 8 bits of input from buffer. */ |
144 | bd->inbufBits = (bd->inbufBits << 8) | bd->inbuf[bd->inbufPos++]; |
145 | bit_count += 8; |
146 | } |
147 | |
148 | /* Calculate result */ |
149 | bit_count -= bits_wanted; |
150 | bd->inbufBitCount = bit_count; |
151 | bits |= (bd->inbufBits >> bit_count) & ((1 << bits_wanted) - 1); |
152 | |
153 | return bits; |
154 | } |
155 | |
156 | /* Unpacks the next block and sets up for the inverse Burrows-Wheeler step. */ |
157 | static int get_next_block(bunzip_data *bd) |
158 | { |
159 | struct group_data *hufGroup; |
160 | int dbufCount, dbufSize, groupCount, *base, *limit, selector, |
161 | i, j, t, runPos, symCount, symTotal, nSelectors, byteCount[256]; |
162 | int runCnt = runCnt; /* for compiler */ |
163 | uint8_t uc, symToByte[256], mtfSymbol[256], *selectors; |
164 | uint32_t *dbuf; |
165 | unsigned origPtr; |
166 | |
167 | dbuf = bd->dbuf; |
168 | dbufSize = bd->dbufSize; |
169 | selectors = bd->selectors; |
170 | |
171 | /* In bbox, we are ok with aborting through setjmp which is set up in start_bunzip */ |
172 | #if 0 |
173 | /* Reset longjmp I/O error handling */ |
174 | i = setjmp(bd->jmpbuf); |
175 | if (i) return i; |
176 | #endif |
177 | |
178 | /* Read in header signature and CRC, then validate signature. |
179 | (last block signature means CRC is for whole file, return now) */ |
180 | i = get_bits(bd, 24); |
181 | j = get_bits(bd, 24); |
182 | bd->headerCRC = get_bits(bd, 32); |
183 | if ((i == 0x177245) && (j == 0x385090)) return RETVAL_LAST_BLOCK; |
184 | if ((i != 0x314159) || (j != 0x265359)) return RETVAL_NOT_BZIP_DATA; |
185 | |
186 | /* We can add support for blockRandomised if anybody complains. There was |
187 | some code for this in busybox 1.0.0-pre3, but nobody ever noticed that |
188 | it didn't actually work. */ |
189 | if (get_bits(bd, 1)) return RETVAL_OBSOLETE_INPUT; |
190 | origPtr = get_bits(bd, 24); |
191 | if ((int)origPtr > dbufSize) return RETVAL_DATA_ERROR; |
192 | |
193 | /* mapping table: if some byte values are never used (encoding things |
194 | like ascii text), the compression code removes the gaps to have fewer |
195 | symbols to deal with, and writes a sparse bitfield indicating which |
196 | values were present. We make a translation table to convert the symbols |
197 | back to the corresponding bytes. */ |
198 | symTotal = 0; |
199 | i = 0; |
200 | t = get_bits(bd, 16); |
201 | do { |
202 | if (t & (1 << 15)) { |
203 | unsigned inner_map = get_bits(bd, 16); |
204 | do { |
205 | if (inner_map & (1 << 15)) |
206 | symToByte[symTotal++] = i; |
207 | inner_map <<= 1; |
208 | i++; |
209 | } while (i & 15); |
210 | i -= 16; |
211 | } |
212 | t <<= 1; |
213 | i += 16; |
214 | } while (i < 256); |
215 | |
216 | /* How many different Huffman coding groups does this block use? */ |
217 | groupCount = get_bits(bd, 3); |
218 | if (groupCount < 2 || groupCount > MAX_GROUPS) |
219 | return RETVAL_DATA_ERROR; |
220 | |
221 | /* nSelectors: Every GROUP_SIZE many symbols we select a new Huffman coding |
222 | group. Read in the group selector list, which is stored as MTF encoded |
223 | bit runs. (MTF=Move To Front, as each value is used it's moved to the |
224 | start of the list.) */ |
225 | for (i = 0; i < groupCount; i++) |
226 | mtfSymbol[i] = i; |
227 | nSelectors = get_bits(bd, 15); |
228 | if (!nSelectors) |
229 | return RETVAL_DATA_ERROR; |
230 | for (i = 0; i < nSelectors; i++) { |
231 | uint8_t tmp_byte; |
232 | /* Get next value */ |
233 | int n = 0; |
234 | while (get_bits(bd, 1)) { |
235 | if (n >= groupCount) return RETVAL_DATA_ERROR; |
236 | n++; |
237 | } |
238 | /* Decode MTF to get the next selector */ |
239 | tmp_byte = mtfSymbol[n]; |
240 | while (--n >= 0) |
241 | mtfSymbol[n + 1] = mtfSymbol[n]; |
242 | mtfSymbol[0] = selectors[i] = tmp_byte; |
243 | } |
244 | |
245 | /* Read the Huffman coding tables for each group, which code for symTotal |
246 | literal symbols, plus two run symbols (RUNA, RUNB) */ |
247 | symCount = symTotal + 2; |
248 | for (j = 0; j < groupCount; j++) { |
249 | uint8_t length[MAX_SYMBOLS]; |
250 | /* 8 bits is ALMOST enough for temp[], see below */ |
251 | unsigned temp[MAX_HUFCODE_BITS+1]; |
252 | int minLen, maxLen, pp, len_m1; |
253 | |
254 | /* Read Huffman code lengths for each symbol. They're stored in |
255 | a way similar to mtf; record a starting value for the first symbol, |
256 | and an offset from the previous value for every symbol after that. |
257 | (Subtracting 1 before the loop and then adding it back at the end is |
258 | an optimization that makes the test inside the loop simpler: symbol |
259 | length 0 becomes negative, so an unsigned inequality catches it.) */ |
260 | len_m1 = get_bits(bd, 5) - 1; |
261 | for (i = 0; i < symCount; i++) { |
262 | for (;;) { |
263 | int two_bits; |
264 | if ((unsigned)len_m1 > (MAX_HUFCODE_BITS-1)) |
265 | return RETVAL_DATA_ERROR; |
266 | |
267 | /* If first bit is 0, stop. Else second bit indicates whether |
268 | to increment or decrement the value. Optimization: grab 2 |
269 | bits and unget the second if the first was 0. */ |
270 | two_bits = get_bits(bd, 2); |
271 | if (two_bits < 2) { |
272 | bd->inbufBitCount++; |
273 | break; |
274 | } |
275 | |
276 | /* Add one if second bit 1, else subtract 1. Avoids if/else */ |
277 | len_m1 += (((two_bits+1) & 2) - 1); |
278 | } |
279 | |
280 | /* Correct for the initial -1, to get the final symbol length */ |
281 | length[i] = len_m1 + 1; |
282 | } |
283 | |
284 | /* Find largest and smallest lengths in this group */ |
285 | minLen = maxLen = length[0]; |
286 | for (i = 1; i < symCount; i++) { |
287 | if (length[i] > maxLen) maxLen = length[i]; |
288 | else if (length[i] < minLen) minLen = length[i]; |
289 | } |
290 | |
291 | /* Calculate permute[], base[], and limit[] tables from length[]. |
292 | * |
293 | * permute[] is the lookup table for converting Huffman coded symbols |
294 | * into decoded symbols. base[] is the amount to subtract from the |
295 | * value of a Huffman symbol of a given length when using permute[]. |
296 | * |
297 | * limit[] indicates the largest numerical value a symbol with a given |
298 | * number of bits can have. This is how the Huffman codes can vary in |
299 | * length: each code with a value>limit[length] needs another bit. |
300 | */ |
301 | hufGroup = bd->groups + j; |
302 | hufGroup->minLen = minLen; |
303 | hufGroup->maxLen = maxLen; |
304 | |
305 | /* Note that minLen can't be smaller than 1, so we adjust the base |
306 | and limit array pointers so we're not always wasting the first |
307 | entry. We do this again when using them (during symbol decoding). */ |
308 | base = hufGroup->base - 1; |
309 | limit = hufGroup->limit - 1; |
310 | |
311 | /* Calculate permute[]. Concurently, initialize temp[] and limit[]. */ |
312 | pp = 0; |
313 | for (i = minLen; i <= maxLen; i++) { |
314 | int k; |
315 | temp[i] = limit[i] = 0; |
316 | for (k = 0; k < symCount; k++) |
317 | if (length[k] == i) |
318 | hufGroup->permute[pp++] = k; |
319 | } |
320 | |
321 | /* Count symbols coded for at each bit length */ |
322 | /* NB: in pathological cases, temp[8] can end ip being 256. |
323 | * That's why uint8_t is too small for temp[]. */ |
324 | for (i = 0; i < symCount; i++) temp[length[i]]++; |
325 | |
326 | /* Calculate limit[] (the largest symbol-coding value at each bit |
327 | * length, which is (previous limit<<1)+symbols at this level), and |
328 | * base[] (number of symbols to ignore at each bit length, which is |
329 | * limit minus the cumulative count of symbols coded for already). */ |
330 | pp = t = 0; |
331 | for (i = minLen; i < maxLen;) { |
332 | unsigned temp_i = temp[i]; |
333 | |
334 | pp += temp_i; |
335 | |
336 | /* We read the largest possible symbol size and then unget bits |
337 | after determining how many we need, and those extra bits could |
338 | be set to anything. (They're noise from future symbols.) At |
339 | each level we're really only interested in the first few bits, |
340 | so here we set all the trailing to-be-ignored bits to 1 so they |
341 | don't affect the value>limit[length] comparison. */ |
342 | limit[i] = (pp << (maxLen - i)) - 1; |
343 | pp <<= 1; |
344 | t += temp_i; |
345 | base[++i] = pp - t; |
346 | } |
347 | limit[maxLen] = pp + temp[maxLen] - 1; |
348 | limit[maxLen+1] = INT_MAX; /* Sentinel value for reading next sym. */ |
349 | base[minLen] = 0; |
350 | } |
351 | |
352 | /* We've finished reading and digesting the block header. Now read this |
353 | block's Huffman coded symbols from the file and undo the Huffman coding |
354 | and run length encoding, saving the result into dbuf[dbufCount++] = uc */ |
355 | |
356 | /* Initialize symbol occurrence counters and symbol Move To Front table */ |
357 | /*memset(byteCount, 0, sizeof(byteCount)); - smaller, but slower */ |
358 | for (i = 0; i < 256; i++) { |
359 | byteCount[i] = 0; |
360 | mtfSymbol[i] = (uint8_t)i; |
361 | } |
362 | |
363 | /* Loop through compressed symbols. */ |
364 | |
365 | runPos = dbufCount = selector = 0; |
366 | for (;;) { |
367 | int nextSym; |
368 | |
369 | /* Fetch next Huffman coding group from list. */ |
370 | symCount = GROUP_SIZE - 1; |
371 | if (selector >= nSelectors) return RETVAL_DATA_ERROR; |
372 | hufGroup = bd->groups + selectors[selector++]; |
373 | base = hufGroup->base - 1; |
374 | limit = hufGroup->limit - 1; |
375 | |
376 | continue_this_group: |
377 | /* Read next Huffman-coded symbol. */ |
378 | |
379 | /* Note: It is far cheaper to read maxLen bits and back up than it is |
380 | to read minLen bits and then add additional bit at a time, testing |
381 | as we go. Because there is a trailing last block (with file CRC), |
382 | there is no danger of the overread causing an unexpected EOF for a |
383 | valid compressed file. |
384 | */ |
385 | if (1) { |
386 | /* As a further optimization, we do the read inline |
387 | (falling back to a call to get_bits if the buffer runs dry). |
388 | */ |
389 | int new_cnt; |
390 | while ((new_cnt = bd->inbufBitCount - hufGroup->maxLen) < 0) { |
391 | /* bd->inbufBitCount < hufGroup->maxLen */ |
392 | if (bd->inbufPos == bd->inbufCount) { |
393 | nextSym = get_bits(bd, hufGroup->maxLen); |
394 | goto got_huff_bits; |
395 | } |
396 | bd->inbufBits = (bd->inbufBits << 8) | bd->inbuf[bd->inbufPos++]; |
397 | bd->inbufBitCount += 8; |
398 | }; |
399 | bd->inbufBitCount = new_cnt; /* "bd->inbufBitCount -= hufGroup->maxLen;" */ |
400 | nextSym = (bd->inbufBits >> new_cnt) & ((1 << hufGroup->maxLen) - 1); |
401 | got_huff_bits: ; |
402 | } else { /* unoptimized equivalent */ |
403 | nextSym = get_bits(bd, hufGroup->maxLen); |
404 | } |
405 | /* Figure how many bits are in next symbol and unget extras */ |
406 | i = hufGroup->minLen; |
407 | while (nextSym > limit[i]) ++i; |
408 | j = hufGroup->maxLen - i; |
409 | if (j < 0) |
410 | return RETVAL_DATA_ERROR; |
411 | bd->inbufBitCount += j; |
412 | |
413 | /* Huffman decode value to get nextSym (with bounds checking) */ |
414 | nextSym = (nextSym >> j) - base[i]; |
415 | if ((unsigned)nextSym >= MAX_SYMBOLS) |
416 | return RETVAL_DATA_ERROR; |
417 | nextSym = hufGroup->permute[nextSym]; |
418 | |
419 | /* We have now decoded the symbol, which indicates either a new literal |
420 | byte, or a repeated run of the most recent literal byte. First, |
421 | check if nextSym indicates a repeated run, and if so loop collecting |
422 | how many times to repeat the last literal. */ |
423 | if ((unsigned)nextSym <= SYMBOL_RUNB) { /* RUNA or RUNB */ |
424 | |
425 | /* If this is the start of a new run, zero out counter */ |
426 | if (runPos == 0) { |
427 | runPos = 1; |
428 | runCnt = 0; |
429 | } |
430 | |
431 | /* Neat trick that saves 1 symbol: instead of or-ing 0 or 1 at |
432 | each bit position, add 1 or 2 instead. For example, |
433 | 1011 is 1<<0 + 1<<1 + 2<<2. 1010 is 2<<0 + 2<<1 + 1<<2. |
434 | You can make any bit pattern that way using 1 less symbol than |
435 | the basic or 0/1 method (except all bits 0, which would use no |
436 | symbols, but a run of length 0 doesn't mean anything in this |
437 | context). Thus space is saved. */ |
438 | runCnt += (runPos << nextSym); /* +runPos if RUNA; +2*runPos if RUNB */ |
439 | if (runPos < dbufSize) runPos <<= 1; |
440 | goto end_of_huffman_loop; |
441 | } |
442 | |
443 | /* When we hit the first non-run symbol after a run, we now know |
444 | how many times to repeat the last literal, so append that many |
445 | copies to our buffer of decoded symbols (dbuf) now. (The last |
446 | literal used is the one at the head of the mtfSymbol array.) */ |
447 | if (runPos != 0) { |
448 | uint8_t tmp_byte; |
449 | if (dbufCount + runCnt > dbufSize) { |
450 | dbg("dbufCount:%d+runCnt:%d %d > dbufSize:%d RETVAL_DATA_ERROR", |
451 | dbufCount, runCnt, dbufCount + runCnt, dbufSize); |
452 | return RETVAL_DATA_ERROR; |
453 | } |
454 | tmp_byte = symToByte[mtfSymbol[0]]; |
455 | byteCount[tmp_byte] += runCnt; |
456 | while (--runCnt >= 0) dbuf[dbufCount++] = (uint32_t)tmp_byte; |
457 | runPos = 0; |
458 | } |
459 | |
460 | /* Is this the terminating symbol? */ |
461 | if (nextSym > symTotal) break; |
462 | |
463 | /* At this point, nextSym indicates a new literal character. Subtract |
464 | one to get the position in the MTF array at which this literal is |
465 | currently to be found. (Note that the result can't be -1 or 0, |
466 | because 0 and 1 are RUNA and RUNB. But another instance of the |
467 | first symbol in the mtf array, position 0, would have been handled |
468 | as part of a run above. Therefore 1 unused mtf position minus |
469 | 2 non-literal nextSym values equals -1.) */ |
470 | if (dbufCount >= dbufSize) return RETVAL_DATA_ERROR; |
471 | i = nextSym - 1; |
472 | uc = mtfSymbol[i]; |
473 | |
474 | /* Adjust the MTF array. Since we typically expect to move only a |
475 | * small number of symbols, and are bound by 256 in any case, using |
476 | * memmove here would typically be bigger and slower due to function |
477 | * call overhead and other assorted setup costs. */ |
478 | do { |
479 | mtfSymbol[i] = mtfSymbol[i-1]; |
480 | } while (--i); |
481 | mtfSymbol[0] = uc; |
482 | uc = symToByte[uc]; |
483 | |
484 | /* We have our literal byte. Save it into dbuf. */ |
485 | byteCount[uc]++; |
486 | dbuf[dbufCount++] = (uint32_t)uc; |
487 | |
488 | /* Skip group initialization if we're not done with this group. Done |
489 | * this way to avoid compiler warning. */ |
490 | end_of_huffman_loop: |
491 | if (--symCount >= 0) goto continue_this_group; |
492 | } |
493 | |
494 | /* At this point, we've read all the Huffman-coded symbols (and repeated |
495 | runs) for this block from the input stream, and decoded them into the |
496 | intermediate buffer. There are dbufCount many decoded bytes in dbuf[]. |
497 | Now undo the Burrows-Wheeler transform on dbuf. |
498 | See http://dogma.net/markn/articles/bwt/bwt.htm |
499 | */ |
500 | |
501 | /* Turn byteCount into cumulative occurrence counts of 0 to n-1. */ |
502 | j = 0; |
503 | for (i = 0; i < 256; i++) { |
504 | int tmp_count = j + byteCount[i]; |
505 | byteCount[i] = j; |
506 | j = tmp_count; |
507 | } |
508 | |
509 | /* Figure out what order dbuf would be in if we sorted it. */ |
510 | for (i = 0; i < dbufCount; i++) { |
511 | uint8_t tmp_byte = (uint8_t)dbuf[i]; |
512 | int tmp_count = byteCount[tmp_byte]; |
513 | dbuf[tmp_count] |= (i << 8); |
514 | byteCount[tmp_byte] = tmp_count + 1; |
515 | } |
516 | |
517 | /* Decode first byte by hand to initialize "previous" byte. Note that it |
518 | doesn't get output, and if the first three characters are identical |
519 | it doesn't qualify as a run (hence writeRunCountdown=5). */ |
520 | if (dbufCount) { |
521 | uint32_t tmp; |
522 | if ((int)origPtr >= dbufCount) return RETVAL_DATA_ERROR; |
523 | tmp = dbuf[origPtr]; |
524 | bd->writeCurrent = (uint8_t)tmp; |
525 | bd->writePos = (tmp >> 8); |
526 | bd->writeRunCountdown = 5; |
527 | } |
528 | bd->writeCount = dbufCount; |
529 | |
530 | return RETVAL_OK; |
531 | } |
532 | |
533 | /* Undo Burrows-Wheeler transform on intermediate buffer to produce output. |
534 | If start_bunzip was initialized with out_fd=-1, then up to len bytes of |
535 | data are written to outbuf. Return value is number of bytes written or |
536 | error (all errors are negative numbers). If out_fd!=-1, outbuf and len |
537 | are ignored, data is written to out_fd and return is RETVAL_OK or error. |
538 | |
539 | NB: read_bunzip returns < 0 on error, or the number of *unfilled* bytes |
540 | in outbuf. IOW: on EOF returns len ("all bytes are not filled"), not 0. |
541 | (Why? This allows to get rid of one local variable) |
542 | */ |
543 | int FAST_FUNC read_bunzip(bunzip_data *bd, char *outbuf, int len) |
544 | { |
545 | const uint32_t *dbuf; |
546 | int pos, current, previous; |
547 | uint32_t CRC; |
548 | |
549 | /* If we already have error/end indicator, return it */ |
550 | if (bd->writeCount < 0) |
551 | return bd->writeCount; |
552 | |
553 | dbuf = bd->dbuf; |
554 | |
555 | /* Register-cached state (hopefully): */ |
556 | pos = bd->writePos; |
557 | current = bd->writeCurrent; |
558 | CRC = bd->writeCRC; /* small loss on x86-32 (not enough regs), win on x86-64 */ |
559 | |
560 | /* We will always have pending decoded data to write into the output |
561 | buffer unless this is the very first call (in which case we haven't |
562 | Huffman-decoded a block into the intermediate buffer yet). */ |
563 | if (bd->writeCopies) { |
564 | |
565 | dec_writeCopies: |
566 | /* Inside the loop, writeCopies means extra copies (beyond 1) */ |
567 | --bd->writeCopies; |
568 | |
569 | /* Loop outputting bytes */ |
570 | for (;;) { |
571 | |
572 | /* If the output buffer is full, save cached state and return */ |
573 | if (--len < 0) { |
574 | /* Unlikely branch. |
575 | * Use of "goto" instead of keeping code here |
576 | * helps compiler to realize this. */ |
577 | goto outbuf_full; |
578 | } |
579 | |
580 | /* Write next byte into output buffer, updating CRC */ |
581 | *outbuf++ = current; |
582 | CRC = (CRC << 8) ^ bd->crc32Table[(CRC >> 24) ^ current]; |
583 | |
584 | /* Loop now if we're outputting multiple copies of this byte */ |
585 | if (bd->writeCopies) { |
586 | /* Unlikely branch */ |
587 | /*--bd->writeCopies;*/ |
588 | /*continue;*/ |
589 | /* Same, but (ab)using other existing --writeCopies operation |
590 | * (and this if() compiles into just test+branch pair): */ |
591 | goto dec_writeCopies; |
592 | } |
593 | decode_next_byte: |
594 | if (--bd->writeCount < 0) |
595 | break; /* input block is fully consumed, need next one */ |
596 | |
597 | /* Follow sequence vector to undo Burrows-Wheeler transform */ |
598 | previous = current; |
599 | pos = dbuf[pos]; |
600 | current = (uint8_t)pos; |
601 | pos >>= 8; |
602 | |
603 | /* After 3 consecutive copies of the same byte, the 4th |
604 | * is a repeat count. We count down from 4 instead |
605 | * of counting up because testing for non-zero is faster */ |
606 | if (--bd->writeRunCountdown != 0) { |
607 | if (current != previous) |
608 | bd->writeRunCountdown = 4; |
609 | } else { |
610 | /* Unlikely branch */ |
611 | /* We have a repeated run, this byte indicates the count */ |
612 | bd->writeCopies = current; |
613 | current = previous; |
614 | bd->writeRunCountdown = 5; |
615 | |
616 | /* Sometimes there are just 3 bytes (run length 0) */ |
617 | if (!bd->writeCopies) goto decode_next_byte; |
618 | |
619 | /* Subtract the 1 copy we'd output anyway to get extras */ |
620 | --bd->writeCopies; |
621 | } |
622 | } /* for(;;) */ |
623 | |
624 | /* Decompression of this input block completed successfully */ |
625 | bd->writeCRC = CRC = ~CRC; |
626 | bd->totalCRC = ((bd->totalCRC << 1) | (bd->totalCRC >> 31)) ^ CRC; |
627 | |
628 | /* If this block had a CRC error, force file level CRC error */ |
629 | if (CRC != bd->headerCRC) { |
630 | bd->totalCRC = bd->headerCRC + 1; |
631 | return RETVAL_LAST_BLOCK; |
632 | } |
633 | } |
634 | |
635 | /* Refill the intermediate buffer by Huffman-decoding next block of input */ |
636 | { |
637 | int r = get_next_block(bd); |
638 | if (r) { /* error/end */ |
639 | bd->writeCount = r; |
640 | return (r != RETVAL_LAST_BLOCK) ? r : len; |
641 | } |
642 | } |
643 | |
644 | CRC = ~0; |
645 | pos = bd->writePos; |
646 | current = bd->writeCurrent; |
647 | goto decode_next_byte; |
648 | |
649 | outbuf_full: |
650 | /* Output buffer is full, save cached state and return */ |
651 | bd->writePos = pos; |
652 | bd->writeCurrent = current; |
653 | bd->writeCRC = CRC; |
654 | |
655 | bd->writeCopies++; |
656 | |
657 | return 0; |
658 | } |
659 | |
660 | /* Allocate the structure, read file header. If in_fd==-1, inbuf must contain |
661 | a complete bunzip file (len bytes long). If in_fd!=-1, inbuf and len are |
662 | ignored, and data is read from file handle into temporary buffer. */ |
663 | |
664 | /* Because bunzip2 is used for help text unpacking, and because bb_show_usage() |
665 | should work for NOFORK applets too, we must be extremely careful to not leak |
666 | any allocations! */ |
667 | int FAST_FUNC start_bunzip(bunzip_data **bdp, int in_fd, |
668 | const void *inbuf, int len) |
669 | { |
670 | bunzip_data *bd; |
671 | unsigned i; |
672 | enum { |
673 | BZh0 = ('B' << 24) + ('Z' << 16) + ('h' << 8) + '0', |
674 | h0 = ('h' << 8) + '0', |
675 | }; |
676 | |
677 | /* Figure out how much data to allocate */ |
678 | i = sizeof(bunzip_data); |
679 | if (in_fd != -1) i += IOBUF_SIZE; |
680 | |
681 | /* Allocate bunzip_data. Most fields initialize to zero. */ |
682 | bd = *bdp = xzalloc(i); |
683 | |
684 | /* Setup input buffer */ |
685 | bd->in_fd = in_fd; |
686 | if (-1 == in_fd) { |
687 | /* in this case, bd->inbuf is read-only */ |
688 | bd->inbuf = (void*)inbuf; /* cast away const-ness */ |
689 | } else { |
690 | bd->inbuf = (uint8_t*)(bd + 1); |
691 | memcpy(bd->inbuf, inbuf, len); |
692 | } |
693 | bd->inbufCount = len; |
694 | |
695 | /* Init the CRC32 table (big endian) */ |
696 | crc32_filltable(bd->crc32Table, 1); |
697 | |
698 | /* Setup for I/O error handling via longjmp */ |
699 | i = setjmp(bd->jmpbuf); |
700 | if (i) return i; |
701 | |
702 | /* Ensure that file starts with "BZh['1'-'9']." */ |
703 | /* Update: now caller verifies 1st two bytes, makes .gz/.bz2 |
704 | * integration easier */ |
705 | /* was: */ |
706 | /* i = get_bits(bd, 32); */ |
707 | /* if ((unsigned)(i - BZh0 - 1) >= 9) return RETVAL_NOT_BZIP_DATA; */ |
708 | i = get_bits(bd, 16); |
709 | if ((unsigned)(i - h0 - 1) >= 9) return RETVAL_NOT_BZIP_DATA; |
710 | |
711 | /* Fourth byte (ascii '1'-'9') indicates block size in units of 100k of |
712 | uncompressed data. Allocate intermediate buffer for block. */ |
713 | /* bd->dbufSize = 100000 * (i - BZh0); */ |
714 | bd->dbufSize = 100000 * (i - h0); |
715 | |
716 | /* Cannot use xmalloc - may leak bd in NOFORK case! */ |
717 | bd->dbuf = malloc_or_warn(bd->dbufSize * sizeof(bd->dbuf[0])); |
718 | if (!bd->dbuf) { |
719 | free(bd); |
720 | xfunc_die(); |
721 | } |
722 | return RETVAL_OK; |
723 | } |
724 | |
725 | void FAST_FUNC dealloc_bunzip(bunzip_data *bd) |
726 | { |
727 | free(bd->dbuf); |
728 | free(bd); |
729 | } |
730 | |
731 | |
732 | /* Decompress src_fd to dst_fd. Stops at end of bzip data, not end of file. */ |
733 | IF_DESKTOP(long long) int FAST_FUNC |
734 | unpack_bz2_stream(transformer_state_t *xstate) |
735 | { |
736 | IF_DESKTOP(long long total_written = 0;) |
737 | bunzip_data *bd; |
738 | char *outbuf; |
739 | int i; |
740 | unsigned len; |
741 | |
742 | if (check_signature16(xstate, BZIP2_MAGIC)) |
743 | return -1; |
744 | |
745 | outbuf = xmalloc(IOBUF_SIZE); |
746 | len = 0; |
747 | while (1) { /* "Process one BZ... stream" loop */ |
748 | |
749 | i = start_bunzip(&bd, xstate->src_fd, outbuf + 2, len); |
750 | |
751 | if (i == 0) { |
752 | while (1) { /* "Produce some output bytes" loop */ |
753 | i = read_bunzip(bd, outbuf, IOBUF_SIZE); |
754 | if (i < 0) /* error? */ |
755 | break; |
756 | i = IOBUF_SIZE - i; /* number of bytes produced */ |
757 | if (i == 0) /* EOF? */ |
758 | break; |
759 | if (i != transformer_write(xstate, outbuf, i)) { |
760 | i = RETVAL_SHORT_WRITE; |
761 | goto release_mem; |
762 | } |
763 | IF_DESKTOP(total_written += i;) |
764 | } |
765 | } |
766 | |
767 | if (i != RETVAL_LAST_BLOCK |
768 | /* Observed case when i == RETVAL_OK: |
769 | * "bzcat z.bz2", where "z.bz2" is a bzipped zero-length file |
770 | * (to be exact, z.bz2 is exactly these 14 bytes: |
771 | * 42 5a 68 39 17 72 45 38 50 90 00 00 00 00). |
772 | */ |
773 | && i != RETVAL_OK |
774 | ) { |
775 | bb_error_msg("bunzip error %d", i); |
776 | break; |
777 | } |
778 | if (bd->headerCRC != bd->totalCRC) { |
779 | bb_error_msg("CRC error"); |
780 | break; |
781 | } |
782 | |
783 | /* Successfully unpacked one BZ stream */ |
784 | i = RETVAL_OK; |
785 | |
786 | /* Do we have "BZ..." after last processed byte? |
787 | * pbzip2 (parallelized bzip2) produces such files. |
788 | */ |
789 | len = bd->inbufCount - bd->inbufPos; |
790 | memcpy(outbuf, &bd->inbuf[bd->inbufPos], len); |
791 | if (len < 2) { |
792 | if (safe_read(xstate->src_fd, outbuf + len, 2 - len) != 2 - len) |
793 | break; |
794 | len = 2; |
795 | } |
796 | if (*(uint16_t*)outbuf != BZIP2_MAGIC) /* "BZ"? */ |
797 | break; |
798 | dealloc_bunzip(bd); |
799 | len -= 2; |
800 | } |
801 | |
802 | release_mem: |
803 | dealloc_bunzip(bd); |
804 | free(outbuf); |
805 | |
806 | return i ? i : IF_DESKTOP(total_written) + 0; |
807 | } |
808 | |
809 | #ifdef TESTING |
810 | |
811 | static char *const bunzip_errors[] = { |
812 | NULL, "Bad file checksum", "Not bzip data", |
813 | "Unexpected input EOF", "Unexpected output EOF", "Data error", |
814 | "Out of memory", "Obsolete (pre 0.9.5) bzip format not supported" |
815 | }; |
816 | |
817 | /* Dumb little test thing, decompress stdin to stdout */ |
818 | int main(int argc, char **argv) |
819 | { |
820 | char c; |
821 | |
822 | int i = unpack_bz2_stream(0, 1); |
823 | if (i < 0) |
824 | fprintf(stderr, "%s\n", bunzip_errors[-i]); |
825 | else if (read(STDIN_FILENO, &c, 1)) |
826 | fprintf(stderr, "Trailing garbage ignored\n"); |
827 | return -i; |
828 | } |
829 | #endif |
830 |