1 /* $NetBSD: inftrees.c,v 1.5 2024/09/22 19:12:27 christos Exp $ */ 2 3 /* inftrees.c -- generate Huffman trees for efficient decoding 4 * Copyright (C) 1995-2024 Mark Adler 5 * For conditions of distribution and use, see copyright notice in zlib.h 6 */ 7 8 #include "zutil.h" 9 #include "inftrees.h" 10 11 #define MAXBITS 15 12 13 const char inflate_copyright[] = 14 " inflate 1.3.1 Copyright 1995-2024 Mark Adler "; 15 /* 16 If you use the zlib library in a product, an acknowledgment is welcome 17 in the documentation of your product. If for some reason you cannot 18 include such an acknowledgment, I would appreciate that you keep this 19 copyright string in the executable of your product. 20 */ 21 22 /* 23 Build a set of tables to decode the provided canonical Huffman code. 24 The code lengths are lens[0..codes-1]. The result starts at *table, 25 whose indices are 0..2^bits-1. work is a writable array of at least 26 lens shorts, which is used as a work area. type is the type of code 27 to be generated, CODES, LENS, or DISTS. On return, zero is success, 28 -1 is an invalid code, and +1 means that ENOUGH isn't enough. table 29 on return points to the next available entry's address. bits is the 30 requested root table index bits, and on return it is the actual root 31 table index bits. It will differ if the request is greater than the 32 longest code or if it is less than the shortest code. 33 */ 34 int ZLIB_INTERNAL inflate_table(codetype type, unsigned short FAR *lens, 35 unsigned codes, code FAR * FAR *table, 36 unsigned FAR *bits, unsigned short FAR *work) { 37 unsigned len; /* a code's length in bits */ 38 unsigned sym; /* index of code symbols */ 39 unsigned mmin, mmax; /* minimum and maximum code lengths */ 40 unsigned root; /* number of index bits for root table */ 41 unsigned curr; /* number of index bits for current table */ 42 unsigned drop; /* code bits to drop for sub-table */ 43 int left; /* number of prefix codes available */ 44 unsigned used; /* code entries in table used */ 45 unsigned huff; /* Huffman code */ 46 unsigned incr; /* for incrementing code, index */ 47 unsigned fill; /* index for replicating entries */ 48 unsigned low; /* low bits for current root entry */ 49 unsigned mask; /* mask for low root bits */ 50 code here; /* table entry for duplication */ 51 code FAR *next; /* next available space in table */ 52 const unsigned short FAR *base; /* base value table to use */ 53 const unsigned short FAR *extra; /* extra bits table to use */ 54 unsigned match; /* use base and extra for symbol >= match */ 55 unsigned short count[MAXBITS+1]; /* number of codes of each length */ 56 unsigned short offs[MAXBITS+1]; /* offsets in table for each length */ 57 static const unsigned short lbase[31] = { /* Length codes 257..285 base */ 58 3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, 19, 23, 27, 31, 59 35, 43, 51, 59, 67, 83, 99, 115, 131, 163, 195, 227, 258, 0, 0}; 60 static const unsigned short lext[31] = { /* Length codes 257..285 extra */ 61 16, 16, 16, 16, 16, 16, 16, 16, 17, 17, 17, 17, 18, 18, 18, 18, 62 19, 19, 19, 19, 20, 20, 20, 20, 21, 21, 21, 21, 16, 203, 77}; 63 static const unsigned short dbase[32] = { /* Distance codes 0..29 base */ 64 1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193, 65 257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145, 66 8193, 12289, 16385, 24577, 0, 0}; 67 static const unsigned short dext[32] = { /* Distance codes 0..29 extra */ 68 16, 16, 16, 16, 17, 17, 18, 18, 19, 19, 20, 20, 21, 21, 22, 22, 69 23, 23, 24, 24, 25, 25, 26, 26, 27, 27, 70 28, 28, 29, 29, 64, 64}; 71 72 /* 73 Process a set of code lengths to create a canonical Huffman code. The 74 code lengths are lens[0..codes-1]. Each length corresponds to the 75 symbols 0..codes-1. The Huffman code is generated by first sorting the 76 symbols by length from short to long, and retaining the symbol order 77 for codes with equal lengths. Then the code starts with all zero bits 78 for the first code of the shortest length, and the codes are integer 79 increments for the same length, and zeros are appended as the length 80 increases. For the deflate format, these bits are stored backwards 81 from their more natural integer increment ordering, and so when the 82 decoding tables are built in the large loop below, the integer codes 83 are incremented backwards. 84 85 This routine assumes, but does not check, that all of the entries in 86 lens[] are in the range 0..MAXBITS. The caller must assure this. 87 1..MAXBITS is interpreted as that code length. zero means that that 88 symbol does not occur in this code. 89 90 The codes are sorted by computing a count of codes for each length, 91 creating from that a table of starting indices for each length in the 92 sorted table, and then entering the symbols in order in the sorted 93 table. The sorted table is work[], with that space being provided by 94 the caller. 95 96 The length counts are used for other purposes as well, i.e. finding 97 the minimum and maximum length codes, determining if there are any 98 codes at all, checking for a valid set of lengths, and looking ahead 99 at length counts to determine sub-table sizes when building the 100 decoding tables. 101 */ 102 103 /* accumulate lengths for codes (assumes lens[] all in 0..MAXBITS) */ 104 for (len = 0; len <= MAXBITS; len++) 105 count[len] = 0; 106 for (sym = 0; sym < codes; sym++) 107 count[lens[sym]]++; 108 109 /* bound code lengths, force root to be within code lengths */ 110 root = *bits; 111 for (mmax = MAXBITS; mmax >= 1; mmax--) 112 if (count[mmax] != 0) break; 113 if (root > mmax) root = mmax; 114 if (mmax == 0) { /* no symbols to code at all */ 115 here.op = (unsigned char)64; /* invalid code marker */ 116 here.bits = (unsigned char)1; 117 here.val = (unsigned short)0; 118 *(*table)++ = here; /* make a table to force an error */ 119 *(*table)++ = here; 120 *bits = 1; 121 return 0; /* no symbols, but wait for decoding to report error */ 122 } 123 for (mmin = 1; mmin <= MAXBITS; mmin++) 124 if (count[mmin] != 0) break; 125 if (root < mmin) root = mmin; 126 127 /* check for an over-subscribed or incomplete set of lengths */ 128 left = 1; 129 for (len = 1; len <= MAXBITS; len++) { 130 left <<= 1; 131 left -= count[len]; 132 if (left < 0) return -1; /* over-subscribed */ 133 } 134 if (left > 0 && (type == CODES || mmax != 1)) 135 return -1; /* incomplete set */ 136 137 /* generate offsets into symbol table for each length for sorting */ 138 offs[1] = 0; 139 for (len = 1; len < MAXBITS; len++) 140 offs[len + 1] = offs[len] + count[len]; 141 142 /* sort symbols by length, by symbol order within each length */ 143 for (sym = 0; sym < codes; sym++) 144 if (lens[sym] != 0) work[offs[lens[sym]]++] = (unsigned short)sym; 145 146 /* 147 Create and fill in decoding tables. In this loop, the table being 148 filled is at next and has curr index bits. The code being used is huff 149 with length len. That code is converted to an index by dropping drop 150 bits off of the bottom. For codes where len is less than drop + curr, 151 those top drop + curr - len bits are incremented through all values to 152 fill the table with replicated entries. 153 154 root is the number of index bits for the root table. When len exceeds 155 root, sub-tables are created pointed to by the root entry with an index 156 of the low root bits of huff. This is saved in low to check for when a 157 new sub-table should be started. drop is zero when the root table is 158 being filled, and drop is root when sub-tables are being filled. 159 160 When a new sub-table is needed, it is necessary to look ahead in the 161 code lengths to determine what size sub-table is needed. The length 162 counts are used for this, and so count[] is decremented as codes are 163 entered in the tables. 164 165 used keeps track of how many table entries have been allocated from the 166 provided *table space. It is checked for LENS and DIST tables against 167 the constants ENOUGH_LENS and ENOUGH_DISTS to guard against changes in 168 the initial root table size constants. See the comments in inftrees.h 169 for more information. 170 171 sym increments through all symbols, and the loop terminates when 172 all codes of length mmax, i.e. all codes, have been processed. This 173 routine permits incomplete codes, so another loop after this one fills 174 in the rest of the decoding tables with invalid code markers. 175 */ 176 177 /* set up for code type */ 178 switch (type) { 179 case CODES: 180 base = extra = work; /* dummy value--not used */ 181 match = 20; 182 break; 183 case LENS: 184 base = lbase; 185 extra = lext; 186 match = 257; 187 break; 188 default: /* DISTS */ 189 base = dbase; 190 extra = dext; 191 match = 0; 192 } 193 194 /* initialize state for loop */ 195 huff = 0; /* starting code */ 196 sym = 0; /* starting code symbol */ 197 len = mmin; /* starting code length */ 198 next = *table; /* current table to fill in */ 199 curr = root; /* current table index bits */ 200 drop = 0; /* current bits to drop from code for index */ 201 low = (unsigned)(-1); /* trigger new sub-table when len > root */ 202 used = 1U << root; /* use root table entries */ 203 mask = used - 1; /* mask for comparing low */ 204 205 /* check available table space */ 206 if ((type == LENS && used > ENOUGH_LENS) || 207 (type == DISTS && used > ENOUGH_DISTS)) 208 return 1; 209 210 /* process all codes and make table entries */ 211 for (;;) { 212 /* create table entry */ 213 here.bits = (unsigned char)(len - drop); 214 if (work[sym] + 1U < match) { 215 here.op = (unsigned char)0; 216 here.val = work[sym]; 217 } 218 else if (work[sym] >= match) { 219 here.op = (unsigned char)(extra[work[sym] - match]); 220 here.val = base[work[sym] - match]; 221 } 222 else { 223 here.op = (unsigned char)(32 + 64); /* end of block */ 224 here.val = 0; 225 } 226 227 /* replicate for those indices with low len bits equal to huff */ 228 incr = 1U << (len - drop); 229 fill = 1U << curr; 230 mmin = fill; /* save offset to next table */ 231 do { 232 fill -= incr; 233 next[(huff >> drop) + fill] = here; 234 } while (fill != 0); 235 236 /* backwards increment the len-bit code huff */ 237 incr = 1U << (len - 1); 238 while (huff & incr) 239 incr >>= 1; 240 if (incr != 0) { 241 huff &= incr - 1; 242 huff += incr; 243 } 244 else 245 huff = 0; 246 247 /* go to next symbol, update count, len */ 248 sym++; 249 if (--(count[len]) == 0) { 250 if (len == mmax) break; 251 len = lens[work[sym]]; 252 } 253 254 /* create new sub-table if needed */ 255 if (len > root && (huff & mask) != low) { 256 /* if first time, transition to sub-tables */ 257 if (drop == 0) 258 drop = root; 259 260 /* increment past last table */ 261 next += mmin; /* here mmin is 1 << curr */ 262 263 /* determine length of next table */ 264 curr = len - drop; 265 left = (int)(1 << curr); 266 while (curr + drop < mmax) { 267 left -= count[curr + drop]; 268 if (left <= 0) break; 269 curr++; 270 left <<= 1; 271 } 272 273 /* check for enough space */ 274 used += 1U << curr; 275 if ((type == LENS && used > ENOUGH_LENS) || 276 (type == DISTS && used > ENOUGH_DISTS)) 277 return 1; 278 279 /* point entry in root table to sub-table */ 280 low = huff & mask; 281 (*table)[low].op = (unsigned char)curr; 282 (*table)[low].bits = (unsigned char)root; 283 (*table)[low].val = (unsigned short)(next - *table); 284 } 285 } 286 287 /* fill in remaining table entry if code is incomplete (guaranteed to have 288 at most one remaining entry, since if the code is incomplete, the 289 maximum code length that was allowed to get this far is one bit) */ 290 if (huff != 0) { 291 here.op = (unsigned char)64; /* invalid code marker */ 292 here.bits = (unsigned char)(len - drop); 293 here.val = (unsigned short)0; 294 next[huff] = here; 295 } 296 297 /* set return parameters */ 298 *table += used; 299 *bits = root; 300 return 0; 301 } 302