inftrees.c revision 1.16
1/* $OpenBSD: inftrees.c,v 1.16 2021/07/04 14:24:49 tb Exp $ */ 2/* inftrees.c -- generate Huffman trees for efficient decoding 3 * Copyright (C) 1995-2017 Mark Adler 4 * For conditions of distribution and use, see copyright notice in zlib.h 5 */ 6 7#include "zutil.h" 8#include "inftrees.h" 9 10#define MAXBITS 15 11 12/* 13 If you use the zlib library in a product, an acknowledgment is welcome 14 in the documentation of your product. If for some reason you cannot 15 include such an acknowledgment, I would appreciate that you keep this 16 copyright string in the executable of your product. 17 */ 18 19/* 20 Build a set of tables to decode the provided canonical Huffman code. 21 The code lengths are lens[0..codes-1]. The result starts at *table, 22 whose indices are 0..2^bits-1. work is a writable array of at least 23 lens shorts, which is used as a work area. type is the type of code 24 to be generated, CODES, LENS, or DISTS. On return, zero is success, 25 -1 is an invalid code, and +1 means that ENOUGH isn't enough. table 26 on return points to the next available entry's address. bits is the 27 requested root table index bits, and on return it is the actual root 28 table index bits. It will differ if the request is greater than the 29 longest code or if it is less than the shortest code. 30 */ 31int ZLIB_INTERNAL inflate_table(type, lens, codes, table, bits, work) 32codetype type; 33unsigned short FAR *lens; 34unsigned codes; 35code FAR * FAR *table; 36unsigned FAR *bits; 37unsigned short FAR *work; 38{ 39 unsigned len; /* a code's length in bits */ 40 unsigned sym; /* index of code symbols */ 41 unsigned min, max; /* minimum and maximum code lengths */ 42 unsigned root; /* number of index bits for root table */ 43 unsigned curr; /* number of index bits for current table */ 44 unsigned drop; /* code bits to drop for sub-table */ 45 int left; /* number of prefix codes available */ 46 unsigned used; /* code entries in table used */ 47 unsigned huff; /* Huffman code */ 48 unsigned incr; /* for incrementing code, index */ 49 unsigned fill; /* index for replicating entries */ 50 unsigned low; /* low bits for current root entry */ 51 unsigned mask; /* mask for low root bits */ 52 code here; /* table entry for duplication */ 53 code FAR *next; /* next available space in table */ 54 const unsigned short FAR *base; /* base value table to use */ 55 const unsigned short FAR *extra; /* extra bits table to use */ 56 unsigned match; /* use base and extra for symbol >= match */ 57 unsigned short count[MAXBITS+1]; /* number of codes of each length */ 58 unsigned short offs[MAXBITS+1]; /* offsets in table for each length */ 59 static const unsigned short lbase[31] = { /* Length codes 257..285 base */ 60 3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, 19, 23, 27, 31, 61 35, 43, 51, 59, 67, 83, 99, 115, 131, 163, 195, 227, 258, 0, 0}; 62 static const unsigned short lext[31] = { /* Length codes 257..285 extra */ 63 16, 16, 16, 16, 16, 16, 16, 16, 17, 17, 17, 17, 18, 18, 18, 18, 64 19, 19, 19, 19, 20, 20, 20, 20, 21, 21, 21, 21, 16, 77, 202}; 65 static const unsigned short dbase[32] = { /* Distance codes 0..29 base */ 66 1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193, 67 257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145, 68 8193, 12289, 16385, 24577, 0, 0}; 69 static const unsigned short dext[32] = { /* Distance codes 0..29 extra */ 70 16, 16, 16, 16, 17, 17, 18, 18, 19, 19, 20, 20, 21, 21, 22, 22, 71 23, 23, 24, 24, 25, 25, 26, 26, 27, 27, 72 28, 28, 29, 29, 64, 64}; 73 74 /* 75 Process a set of code lengths to create a canonical Huffman code. The 76 code lengths are lens[0..codes-1]. Each length corresponds to the 77 symbols 0..codes-1. The Huffman code is generated by first sorting the 78 symbols by length from short to long, and retaining the symbol order 79 for codes with equal lengths. Then the code starts with all zero bits 80 for the first code of the shortest length, and the codes are integer 81 increments for the same length, and zeros are appended as the length 82 increases. For the deflate format, these bits are stored backwards 83 from their more natural integer increment ordering, and so when the 84 decoding tables are built in the large loop below, the integer codes 85 are incremented backwards. 86 87 This routine assumes, but does not check, that all of the entries in 88 lens[] are in the range 0..MAXBITS. The caller must assure this. 89 1..MAXBITS is interpreted as that code length. zero means that that 90 symbol does not occur in this code. 91 92 The codes are sorted by computing a count of codes for each length, 93 creating from that a table of starting indices for each length in the 94 sorted table, and then entering the symbols in order in the sorted 95 table. The sorted table is work[], with that space being provided by 96 the caller. 97 98 The length counts are used for other purposes as well, i.e. finding 99 the minimum and maximum length codes, determining if there are any 100 codes at all, checking for a valid set of lengths, and looking ahead 101 at length counts to determine sub-table sizes when building the 102 decoding tables. 103 */ 104 105 /* accumulate lengths for codes (assumes lens[] all in 0..MAXBITS) */ 106 for (len = 0; len <= MAXBITS; len++) 107 count[len] = 0; 108 for (sym = 0; sym < codes; sym++) 109 count[lens[sym]]++; 110 111 /* bound code lengths, force root to be within code lengths */ 112 root = *bits; 113 for (max = MAXBITS; max >= 1; max--) 114 if (count[max] != 0) break; 115 if (root > max) root = max; 116 if (max == 0) { /* no symbols to code at all */ 117 here.op = (unsigned char)64; /* invalid code marker */ 118 here.bits = (unsigned char)1; 119 here.val = (unsigned short)0; 120 *(*table)++ = here; /* make a table to force an error */ 121 *(*table)++ = here; 122 *bits = 1; 123 return 0; /* no symbols, but wait for decoding to report error */ 124 } 125 for (min = 1; min < max; min++) 126 if (count[min] != 0) break; 127 if (root < min) root = min; 128 129 /* check for an over-subscribed or incomplete set of lengths */ 130 left = 1; 131 for (len = 1; len <= MAXBITS; len++) { 132 left <<= 1; 133 left -= count[len]; 134 if (left < 0) return -1; /* over-subscribed */ 135 } 136 if (left > 0 && (type == CODES || max != 1)) 137 return -1; /* incomplete set */ 138 139 /* generate offsets into symbol table for each length for sorting */ 140 offs[1] = 0; 141 for (len = 1; len < MAXBITS; len++) 142 offs[len + 1] = offs[len] + count[len]; 143 144 /* sort symbols by length, by symbol order within each length */ 145 for (sym = 0; sym < codes; sym++) 146 if (lens[sym] != 0) work[offs[lens[sym]]++] = (unsigned short)sym; 147 148 /* 149 Create and fill in decoding tables. In this loop, the table being 150 filled is at next and has curr index bits. The code being used is huff 151 with length len. That code is converted to an index by dropping drop 152 bits off of the bottom. For codes where len is less than drop + curr, 153 those top drop + curr - len bits are incremented through all values to 154 fill the table with replicated entries. 155 156 root is the number of index bits for the root table. When len exceeds 157 root, sub-tables are created pointed to by the root entry with an index 158 of the low root bits of huff. This is saved in low to check for when a 159 new sub-table should be started. drop is zero when the root table is 160 being filled, and drop is root when sub-tables are being filled. 161 162 When a new sub-table is needed, it is necessary to look ahead in the 163 code lengths to determine what size sub-table is needed. The length 164 counts are used for this, and so count[] is decremented as codes are 165 entered in the tables. 166 167 used keeps track of how many table entries have been allocated from the 168 provided *table space. It is checked for LENS and DIST tables against 169 the constants ENOUGH_LENS and ENOUGH_DISTS to guard against changes in 170 the initial root table size constants. See the comments in inftrees.h 171 for more information. 172 173 sym increments through all symbols, and the loop terminates when 174 all codes of length max, i.e. all codes, have been processed. This 175 routine permits incomplete codes, so another loop after this one fills 176 in the rest of the decoding tables with invalid code markers. 177 */ 178 179 /* set up for code type */ 180 switch (type) { 181 case CODES: 182 base = extra = work; /* dummy value--not used */ 183 match = 20; 184 break; 185 case LENS: 186 base = lbase; 187 extra = lext; 188 match = 257; 189 break; 190 default: /* DISTS */ 191 base = dbase; 192 extra = dext; 193 match = 0; 194 } 195 196 /* initialize state for loop */ 197 huff = 0; /* starting code */ 198 sym = 0; /* starting code symbol */ 199 len = min; /* starting code length */ 200 next = *table; /* current table to fill in */ 201 curr = root; /* current table index bits */ 202 drop = 0; /* current bits to drop from code for index */ 203 low = (unsigned)(-1); /* trigger new sub-table when len > root */ 204 used = 1U << root; /* use root table entries */ 205 mask = used - 1; /* mask for comparing low */ 206 207 /* check available table space */ 208 if ((type == LENS && used > ENOUGH_LENS) || 209 (type == DISTS && used > ENOUGH_DISTS)) 210 return 1; 211 212 /* process all codes and make table entries */ 213 for (;;) { 214 /* create table entry */ 215 here.bits = (unsigned char)(len - drop); 216 if (work[sym] + 1U < match) { 217 here.op = (unsigned char)0; 218 here.val = work[sym]; 219 } 220 else if (work[sym] >= match) { 221 here.op = (unsigned char)(extra[work[sym] - match]); 222 here.val = base[work[sym] - match]; 223 } 224 else { 225 here.op = (unsigned char)(32 + 64); /* end of block */ 226 here.val = 0; 227 } 228 229 /* replicate for those indices with low len bits equal to huff */ 230 incr = 1U << (len - drop); 231 fill = 1U << curr; 232 min = fill; /* save offset to next table */ 233 do { 234 fill -= incr; 235 next[(huff >> drop) + fill] = here; 236 } while (fill != 0); 237 238 /* backwards increment the len-bit code huff */ 239 incr = 1U << (len - 1); 240 while (huff & incr) 241 incr >>= 1; 242 if (incr != 0) { 243 huff &= incr - 1; 244 huff += incr; 245 } 246 else 247 huff = 0; 248 249 /* go to next symbol, update count, len */ 250 sym++; 251 if (--(count[len]) == 0) { 252 if (len == max) break; 253 len = lens[work[sym]]; 254 } 255 256 /* create new sub-table if needed */ 257 if (len > root && (huff & mask) != low) { 258 /* if first time, transition to sub-tables */ 259 if (drop == 0) 260 drop = root; 261 262 /* increment past last table */ 263 next += min; /* here min is 1 << curr */ 264 265 /* determine length of next table */ 266 curr = len - drop; 267 left = (int)(1 << curr); 268 while (curr + drop < max) { 269 left -= count[curr + drop]; 270 if (left <= 0) break; 271 curr++; 272 left <<= 1; 273 } 274 275 /* check for enough space */ 276 used += 1U << curr; 277 if ((type == LENS && used > ENOUGH_LENS) || 278 (type == DISTS && used > ENOUGH_DISTS)) 279 return 1; 280 281 /* point entry in root table to sub-table */ 282 low = huff & mask; 283 (*table)[low].op = (unsigned char)curr; 284 (*table)[low].bits = (unsigned char)root; 285 (*table)[low].val = (unsigned short)(next - *table); 286 } 287 } 288 289 /* fill in remaining table entry if code is incomplete (guaranteed to have 290 at most one remaining entry, since if the code is incomplete, the 291 maximum code length that was allowed to get this far is one bit) */ 292 if (huff != 0) { 293 here.op = (unsigned char)64; /* invalid code marker */ 294 here.bits = (unsigned char)(len - drop); 295 here.val = (unsigned short)0; 296 next[huff] = here; 297 } 298 299 /* set return parameters */ 300 *table += used; 301 *bits = root; 302 return 0; 303} 304